Essays in Corporate Finance and Financial Institutions
by
Adam C. Kolasinski
BA, Economics and Mathematics
Columbia College, Columbia University in the City of New York
Submitted to the Sloan School of Management
in Fulfillment of the Requirements for the Degree of
Doctor of Philosophy in Management
at the
Massachusetts Institute of Technology
June, 2006
Chapters 1 & 2, C Adam C. Kolasinski, all rights reserved.
Chapter 3, C Adam C. Kolasinski and S.P. Kothari, all rights reserved
The author(s) hereby grants to MIT permission to reproduce and distribute publicly paper
and electronic copies of this thesis document in whole or in part in any medium now
known or hereafter created.
Signature of Author:.
ri N
Certified by:
S.P. Kothari, Thesis Committee Chairman
Gordon Y. Billard Professor of Management
Accepted by:
Birg"Wernerfelt, Ph.D. Program Chairman
J.C. Penny Professor of Management Science
OFTECHNOLOGY
AUG 312006
LIBRARIES
I
AMCHIVES
I
Essays in Corporate Finance and Financial Institutions
by
Adam C. Kolasinski
Submitted to the Sloan School of Management in Fulfillment of the Requirements for the
Degree of Doctor of Philosophy in Management
Abstract:
Chi: Subsidiary Debt, Capital Structure, and Internal Capital Markets
I investigate external subsidiary debt financing and its implications for internal capital
markets. I find that firms tend to finance business segments with subsidiary debt when
those segments have better investment opportunities than the rest of the firm, and such
debt tends to be parent-guaranteed. I also find that having such debt outstanding
significantly reduces the effect of a segment's cash flow on the capital expenditures of
other segments. These findings suggest that firms use subsidiary debt to protect their
stronger segments from the underfunding or "poaching" problems modeled in theories of
internal capital markets. In addition, I find that firms use subsidiary debt for reasons
related to traditional capital structure concerns.
Ch2: Is the Chinese Wall too High?
I test whether new regulatory restrictions on cooperation between analysts and
investment bankers adversely affect equity research coverage. Contrary to the
hypothesis, I find that firms engaging in SEO's enjoy just as large an increase in analyst
coverage in the post-regulatory period as they do in the pre-regulatory period. In addition,
while I find that analyst coverage in the post regulatory period significantly declines for
new IPOs, it declines by an equal amount for a control group of comparable firms that
pay no such fees. Making the identifying assumption that any adverse consequences of
the new restrictions should be larger for IPO's, I conclude that the restrictions have no
adverse impact on analyst coverage.
Ch3: Investment Banking and Analyst Objectivity'
This chapter uncovers evidence that conflicts of interest arising from M&A advisory
relations influence amlysts' recommendations, corroborating regulators' and
practitioners' suspicions on a topic not previously examined in the academic literature.
In addition, the M&A context allows us to disentangle the conflict of interest effect from
selection bias. We find that analysts affiliated with acquirer advisors upgrade acquirer
stocks around M&A deals, even around all-cash deals, wherein selection bias is unlikely.
Also consistent with conflict of interest, but not selection bias, target-affiliated analysts
publish optimistic reports about acquirers after, but not before, the exchange ratio of an
all-stock deal is set.
Thesis committee:
S.P. Kothari, Gordan Y. Billard Professor of Management, Chairman
Stewart C. Myers, Gordan Y. Billard Professor of Finance
Antoinette Schoar, Michael M. Koerner Associate Professor of Entrepreneurial Finance
SChapter 3 is joint work with S.P. Kothari
Table of Contents
Chapter 1: Subsidiary Debt, Capital Structure, and Internal Capital Markets ................ 4
Chapter 2: Is the Chinese Wall to High? Investigating the Costs of
New Restrictions on Cooperation Between Analysts and Investment Bankers............53
Chapter 3: Investment Banking and Analyst Objectivity: Evidence
From Analysts Affiliated with M&A advisors .................................................
81
Acknowledgements:
I thank my dissertation committee members, S.P. Kothari, Stewart Myers, and Antoinette Schoar, for their
advice and guidance. I thank Tobias Adrian, Andres Almazan, Paul Asquith, Jack Bao, Utpal
Bhattacharya, Nittai Bergman, John Chalmers, Harry & Linda DeAngelo, Alex Edmans, Mara Faccio,
Paolo Fulghieri, Ilan Guedj, Jarrad Harford, Dirk Jenter, Philippe Jorion, Jiro Kondo, Andrew Lo, Roni
Michaeli, Stas Nikolova, Kevin Rock, Ed Rice, Roberto Rigobon, Jeremy Stein, and Alan Timmerman as
well as seminar participants at the Federal Reserve Bank of New York, George Mason University, Indiana
University, MIT, UC Irvine, UC Sand Diego, USC, the University of Washington, and Vanderbilt
University for their helpful comments and suggestions. I thank Solomon Samson of Standard & Poor's and
Pamela Stumpp of Moody's for providing insights about legal and institutional details. I am grateful to
State Farm Companies for their financial support. Any errors are my own.
Chapter 1
Subsidiary Debt, Capital Structure, and Internal Capital Markets
1. Introduction
This study proposes and tests rationales for why non-financial firms use
subsidiary-level debt financing, a topic that remains largely unexplored in the academic
literature. In addition, I examine how subsidiary debt affects the efficiency of internal
capital markets. Subsidiary debt is a popular form of financing. Since 1995, subsidiary
debt issues have accounted for approximately 13% of total US non-financial corporate
public debt proceeds. Unlike most corporate debt, subsidiary debt is backed not by the
entire firm, but a division organized as a distinct legal entity, i.e. a subsidiary. 2 Holders
of such debt have a claim on the subsidiary senior to that of the parent's creditors. If the
parent guarantees the debt, subsidiary debt holders also have recourse to the parent. 3
Otherwise, they have no recourse, unless the parent engages in some wrong-doing
(Thomson, 1991).
I estimate multinomial logistic models of the probability that a division has
subsidiary debt outstanding, and whether it is guaranteed by the parent. I find that a
division is more likely to have subsidiary debt outstanding if its parent has a junk credit
rating, if it is in a less volatile industry than other segments in the firm, and if the parent
firm's divisions vary more in Tobin's q and industry credit risk. In all the above cases,
the debt tends not be parent guaranteed. In addition, I find that a division is more likely
Project finance debt also allows firms to borrow against subsets of themselves. Subsidiary differs in that
it is backed by a going concern rather than a special purpose vehicle created to undertake a finite-lived
ýroject (Kensigner and Martin, 1988).
If the subsidiary is organized as an unlimited liability entity or a partnership with the parent as general
partner, its debt holders have a claim on parent assets even without a guarantee.
2
to have debt outstanding if it has better investment opportunities than the rest of the firm,
and in this case the debt tends to be parent-guaranteed.
The last result suggests that diversified firms use parent-guaranteed subsidiary
debt to mitigate the internal capital market inefficiencies modeled in Scharfstein and
Stein (S&S 2000) and Rajan, Servaes and Zingales (RSZ 2000). These models postulate
that inefficiencies arise because divisions richest in investment opportunities tend to get
"poached" by other divisions or underfunded by the parent. By financing such divisions
with subsidiary debt, firms can commit to clearly delineating their profits as well as
allocating more capital to them and keeping it there, counteracting the poaching and
underfunding problems. Confirming this notion, I find that divisions with parentguaranteed subsidiary debt outstanding are significantly less likely to have their cash flow
diverted and invested in other divisions.
My results have important implications for the debate over internal capital
markets. Berger and Ofek (1995), Lamont (1997), Lamont and Polk (2002), Lang and
Stulz (1994), Ozbas (2003), Rajan, Servaes and Zingales (2000), Scharfstein (1998), and
Shin and Stulz (1998) find evidence that investment polices of diversified firms appear
less efficient than that of standalones. However, Gomes and Livdan (2004) and
Matsusaka (2001) construct models in which less productive firms with poorer
investment opportunities are more likely to diversify, suggesting that endogeneity could
explain some of the above findings. Campa and Kedia (2002), Chevalier (2000),
Graham, Lemmon and Wolf (2002), Maksimovic and Phillips (2002), and Schoar (2002)
find evidence that the endogeneity of diversification is important. Villalonga (2004) and
Whited (2001) find evidence that measurement error in proxies for investment
opportunities may also explain some of the apparent inefficiency in conglomerate
investment policy. These econometric issues, however, do not affect my tests, which
support the S&S and RSZ theories of inefficient internal capital markets. 4
My finding that firms tend to place subsidiary debt on their least volatile divisions
supports the hypothesis that firms use subsidiary debt to reduce costs associated with
asymmetric information about risk. Previous research has used volatility as a proxy for
information asymmetry about risk (Haddock and James, 2002) as well as risk itself. Thus
my results indicate that firms place non- guaranteed subsidiary debt at divisions where the
costs of asymmetric information about risk are lowest. 5 My finding that junk credit firms
are more likely to use non-guaranteed subsidiary debt also supports the asymmetric
information cost hypothesis since such costs are increasing in risk.
That firms with junk credit are more likely to use subsidiary debt also supports the
hypothesis that firms use such debt to mitigate costs of financial distress, either in the
form of debt overhang (Myers, 1977) or risk-shifting incentives (Jensen and Meckling,
1976). As explained in Section 2, non-guaranteed subsidiary debt placed at one division
can prevent financial distress at that division from inducing either underinvestment or
risk-shifting in others. Such concerns become more salient as the probability of financial
distress increases, as is indicated by a junk credit rating. Further supporting the riskshifting hypothesis, I find that firms whose divisions vary more in industry credit risk
tend to use non- guaranteed subsidiary debt. Supporting the debt overhang hypothesis, I
4 Other theories of inefficient internal capital markets include Harris and Raviv (1996, 1998), Ozbas (2005),
Wulf (2002), and Goel, Nanda and Narayanan (2004). None has any implications for subsidiary debt.
5 My results also indicate that firms issue subsidiary debt at their safest divisions. That a division is safer,
however, does notperse provide a rationale for financing it with subsidiary debt. If having seniority over
the safer division's assets makes a debt issue safer, it also makes the firm's other securities riskier, leaving
the firm's total risk and cost of capital unchanged.
find that firms whose divisions vary more in Tobin's q, a proxy for investment
opportunities, are also more likely to use non-guaranteed subsidiary debt.
I have one additional result: firms that use subsidiary debt tend to buy and sell
subsidiaries more frequently than firms that do not use it. This is consistent with two
hypotheses. First, public subsidiary debt may help increase a subsidiary's divestiture
price by establishing for it a track record of financial performance. Second, subsidiary
debt may reduce divestiture transactions costs. To protect credit quality, parent creditors
often insist on covenants restricting divestitures. Thus, divesting a subsidiary often
requires obtaining waivers, which can be costly due to hold up problems. In contrast,
non-guaranteed subsidiary creditors have no recourse to the parent, making them little
affected by the divestiture and therefore unlikely to hold it up.
Some of my rationales for subsidiary debt are similar to those proposed for equity
carve-outs. Nanda (1991) hypothesizes that carve-outs reduce information asymmetry
costs. Slovin and Sushka (1997) find supporting evidence, but Vijh (2002 and 2005)
finds contrary evidence. In addition, Vijh (2002) asserts that equity carve-outs help
"focus" a firm. As I argue in Section 2, carve-outs might improve internal capital
markets in a manner similar to that of subsidiary debt, so I control for minority equity
stakes in tests where relevant. My results are not sensitive to such controls.
The rest of this study proceeds as follows. In Section 2, I develop rationales for
subsidiary debt use and their empirical implications. In Section 3, I discuss my sample.
In Section 4, I discuss my tests and present results. In Section 5, I examine alternative
rationales for subsidiary debt and rule them out. Section 6 concludes.
2. Hypothesis development
In this section I provide rationales for subsidiary debt based on internal capital
markets and classical capital structure theory. I also develop empirical hypotheses that
these rationales imply, which I summarize in Table 1.
2.1 Subsidiary Debt and Models of Internal CapitalMarkets
Subsidiary debt may improve internal capital markets in two ways. First, with
restrictions on use of proceeds and asset sales, subsidiary debt may allow the CEO to
commit to invest more capital in divisions that have the best investment opportunities in
the firm and keep it there. Thus CEOs can overcome ex-post incentives, modeled in
Scharfstein and Stein (S&S, 2000), to underfund such divisions. Second, through
disclosure requirements, restrictive covenants, and debt service, subsidiary debt may bind
a CEO to prevent divisions with poor investment opportunities from "poaching"
opportunity-rich divisions, as modeled in Rajan, Servaes, and Zingales (RSZ, 2000).
Below, I elaborate on how subsidiary debt fits into the S&S and RSZ models.
S&S assume that managers of investment opportunity-poor divisions engage in
rent-seeking activities, whereas opportunity-rich division managers do not.6 As a result,
the CEO must pay rents to managers of opportunity-poor divisions. The latter are empire
builders, so they will accept rents in the form of an increased capital budget as well as
cash bonuses. The CEO wants to keep excess cash in the bonus pool for himself, so he
pays rents by distorting the capital budget in poor division managers' favor. Anticipating
this problem, investors curtail both the firm's capital budget and bonus pool.
S&S assume that managers face a tradeoff between rent-seeking and productive work. A manager's
return on productive work is higher when his division has better investment opportunities. Hence the
opportunity cost of rent-seeking is lower for opportunity-poor division managers than for opportunity-rich
division managers, prompting only the former to rent-seek.
6
The CEO wants as much capital as possible, and in equilibrium he fails to keep
excess bonus cash because investors curtail the pool. Therefore, if it were possible, the
CEO would prefer to commit ex-ante to invest more in the opportunity-rich division,
which he can achieve by financing the division with subsidiary debt. Provisions on use
of proceeds can bind the CEO to invest the newly- raised capital in the division.
Covenants restricting capital transfers can bind him to keep it there.
Parent securities may also have provisions binding the CEO to invest proceeds in
certain projects. Thus subsidiary debt is not necessary to prevert underfunding of high
quality projects. However, specifying all projects liable to get underfunded may not be
possible. For instance, it may be known that the best projects are in a particular division,
but the precise investments needed to undertake them may not be known or impossible to
contractually specify. In such cases, the firm should commit funds to the division, as
subsidiary debt allows. Separately incorporating the division as a subsidiary and giving it
its own debt holders reduces the CEO's ability to renege on such a commitment.
If the opportunity-poor division is expected to continue having poor investment
opportunities for some time, the CEO will want to commit to not giving it other
divisions' future cash flows. In the spirit of Jensen (1986), issuing subsidiary debt
against those cash flows ensures they get paid out to creditors instead of the opportunity
poor division. Restrictions on use of proceeds can ensure the present value of those cash
flows gets allocated to the opportunity-rich division at time of issuance.
RSZ postulate that managers of opportunity-poor divisions can "poach" the
surplus of opportunity-rich divisions, and the CEO cannot commit to prohibiting it.
Threat of poaching, in turn, induces opportunity-rich divisions to make "defensive"
investments whose surplus is lower but more difficult to poach. As a second-best
solution, the CEO allocates to poor divisions more capital than is efficient. With more
capital, poor division managers derive greater utility from productive work than
poaching, and thus choose not to poach. Opportunity-rich divisions make the poachable,
high surplus investments, albeit with less capital than is efficient.
If it were possible, the CEO would want to commit to prohibit poaching since it
would result in higher total surplus. Financing opportunity-rich divisions with subsidiary
debt allows him to make this commitment. Public subsidiary debt legally obliges the
firm to delineate and disclose the opportunity-rich division's profits, which may reduce
the ability of poor division managers to claim a portion as their own and makes poaching
more visible to outsiders, providing ex post incentives for the CEO to prevent it. In
addition, debt service may shield the opportunity-rich division surplus available to poach,
the present value of the surplus having been allocated to the opportunity-rich division at
time of issuance. Restrictive covenants on dividends to the parent may also help.
Equity carve-outs, which also carry use of proceeds provisions, minority
shareholder protections, and disclosure requirements, may also allow a CEO to commit to
investing in a division and prevent poaching. I focus on subsidiary debt because it is
more common, leaving to future research the internal capital market effect of carve-outs.
Nevertheless, I control for carve-outs and other minority equity stakes where appropriate.
Both the above theories of inefficient internal capital markets imply that
diversified firms will have subsidiary debt at those divisions with better investment
opportunities relative to others. Furthermore, the parent should guarantee this debt for at
least three reasons. First, a parent guarantee will minimize debt overhang problems in the
division, which are likely substantial given that it is rich in investment opportunities.
Should the division enter distress, without a guarantee, the parent has ex-post incentives
to forego its valuable investment opportunities. Second, a parent guarantee can help
alleviate adverse selection problems, which are likely to be high since divisions rich in
investment opportunities are likely more opaque. Finally, a guarantee is a public signal
that the firm is committed to a given division. Thus, empirically, divisions should be
more likely to be financed with subsidiary debt the richer they are in investment
opportunities relative to the rest of the firm.7
If guaranteed subsidiary debt successfully protects the surplus of opportunity-rich
divisions from poaching, cash flow of the indebted division should not get diverted to
other divisions. Although RSZ do not specify what opportunity-poor division managers
do with poached surplus, given the ubiquity of empire-building, it is reasonable to
assume they increase capital expenditures. Using subsidiary debt to prevent poaching,
therefore, has the following empirical implication: when a division has guaranteed
subsidiary debt outstanding, its cash flow should have less effect on other divisions'
investment. This implication also follows if firms use subsidiary debt to mitigate
inefficiencies modeled in S&S. In this case, the CEO uses subsidiary debt to commit to
not investing in other divisions the cash flows generated by the indebted division.
2.2 Subsidiary Debt and ClassicalCapitalStructure Theory: Avoiding Debt Overhang
In his analysis of debt overhang, Myers (1977) shows that diversification brings
the benefit of coinsurance but also introduces the possibility of cross-division financial
contagion. John (1993) extends this analysis. Thus diversification has effects that both
7 The guarantee need not be explicit. Reputation concerns or the division's importance to the firm's other
operations may give the firm sufficient incentives to bail it out in the event of distress. See Stumpp,
Rotino, Fanger, Stephan, and Carter (2003).
increase and decrease the expected costs of debt overhang. As I explain below, a
diversified firm can use non-guaranteed subsidiary debt to reduce the potential for crossdivision contagion while at the same time preserving much of the benefit of coinsurance.
The benefit of coinsurance arises because poor performance at some divisions
may be offset by good performance at others, lowering overall probability of parent
financial distress, and hence the expected costs of debt overhang. If the firm only uses
parent level financing, however, sufficiently poor performance at just one division can
induce healthy divisions to forgo valuable investment opportunities. Instead, if the
division that performs poorly had been financed with non-guaranteed subsidiary debt, the
firm can let it default, insulating healthy divisions from its distress. On the other hand, if
the indebted division performs well, its surplus cash can bail out other divisions should
they perform poorly.
Using non- guaranteed subsidiary debt in this manner carries a cost: should the
indebted division enter financial distress, its valuable investment opportunities will likely
be forfeited even in states where the firm as a whole is solvent. This cost, however, is
low if the indebted division has poor investment opportunities. At the same time, if other
divisions in the firm are rich in quality investment opportunities, there is a large benefit to
insulating them from financial distress at the opportunity-poor division. Thus firms
whose divisions differ greatly in quality of investment opportunities should finance with
non-guaranteed subsidiary debt their opportunity-poor divisions. John and John (1991)
provide a similar rationale for non-recourse project finance debt.
Firms should not place non- guaranteed subsidiary debt at opportunity-rich
divisions. Should such divisions experience a sufficiently adverse cash flow shock, the
firm will have an incentive to let them default and forgo their investment opportunities.
2.3 Subsidiary Debt and ClassicalCapitalStructure Theory: Risk-shifting
The risk-shifting problem first described in Jensen and Meckling (1976) is
another source of financial distress costs. When a firm enters distress, equity holders
have an incentive to substitute existing assets for more risky ones in an attempt to
"gamble for resurrection," even if the new assets have negative NPV. Kahn and Winton
(2004) show that diversified financial institutions can reduce risk-shifting incentives by
financing their riskiest subsidiaries with non-guaranteed debt. Should these subsidiaries
enter distress, their creditors have no claim on safer subsidiaries, reducing equity holders'
incentive for negative NPV risk-shifting in the latter.
Kahn and Winton only apply their model to financial institutions because they
have high leverage and flexible asset composition. However, some non-financial firms
share these characteristics. Flannery, Houston and Venkataraman (1993) argue that
primarily non-financial diversified firms can reduce risk-shifting incentives by financing
with non- guaranteed debt their risky financial subsidiaries. Their argument could also
apply to risky non- financial subsidiaries. Guaranteed subsidiary debt would not be
effective since its holders have recourse to the parent, and by extension the safer division.
Financing safer divisions with subsidiary debt may also prevent a form of riskshifting wherein capital from safer divisions is transferred to riskier divisions. Covenants
on subsidiary debt often restrict such transfers. It might be possible to add similar
covenants to parent-debt issues, obviating the need for subsidiary debt. However, it is
plausible that such covenants are easier to enforce when the safer division is a separately
incorporated legal entity with its own public debt and audited financial statements.
In summary, the risk-shifting rationale predicts that riskier firms whose divisions
differ more in risk will use subsidiary debt at all divisions. The parent should not
guarantee debt is sued at riskier divisions.
2.4 SubsidiaryDebt and ClassicalCapitalStructure Theory: Asymmetric Information
In sections 3.3 and 3.4, Myers and Majluf (1984) argue that when insiders have
better information about firm risk, the decision to issue debt constitutes bad news. Thus
investors in such new issues will demand a discount. If information asymmetry varies
across a firm's divisions, a firm can reduce borrowing costs by issuing debt at divisions
with lowest asymmetry. In addition, costs of information asymmetry about risk are
increasing in risk, so firms should also place debt on their safest divisions. Assuming
volatility proxies for both information asymmetry and risk, the above rationale for
subsidiary debt predicts that firms should issue subsidiary debt at their least volatile
divisions. In addition, firms with junk credit ratings should be more likely to use this
strategy since they are riskiest.
Empirical evidence suggests that there is no relation between firm diversity and
information asymmetry (see Clarke, Fee and Thomas, 2004, Hadlock, Ryngaert, and
Thomas, 2001, and Thomas, 2002). Therefore, the information asymmetry rationale for
subsidiary debt makes no predictions about firm diversity.
For subsidiary debt to limit asymmetric information costs, it must insulate a
subsidiary from parent distress. Otherwise, subsidiary creditors are no less exposed to
parent risk than parent creditors. In fact, distressed parents often can strip sound
subsidiaries. Therefore, subsidiaries rarely have credit ratings higher than their parents,
even if less risky as standalones (Samson, Sprinzen and Dobois-Pelerin, 2005).
Nevertheless, subsidiary debt holders can sue if the parent strips a subsidiary. A bankrupt
parent can often force a sound subsidiary into bankruptcy, but subsidiary creditors' legal
priority lowers their recovery risk (Samson et al., 2005). Thus subsidiary debt provides
some insulation from parent distress.
Note that firms should only use non- guaranteed subsidiary debt to limit
information asymmetry costs since a parent guarantee is a bad signal about parent risk.
2.5 Subsidiary Debt and M&A Costs
Non-guaranteed subsidiary debt can reduce the cost of selling subsidiaries. To
protect debt holders' interests, parent debt often carries provisions restricting the sale of
subsidiaries. Thus firms may have to obtain waivers each time they sell a subsidiary,
which can be costly due to holdup problems. Such a sale, however, does not affect the
holders of non- guaranteed subsidiary debt since they have no recourse to the parent.
Subsidiaries with public debt are obliged to publicly report financial results, a history of
which might reduce information asymmetry and thereby help the parent get a better
divestiture price. Hence firms may issue public subsidiary debt in preparation for a
divestiture. Thus firms that use subsidiary debt likely transact in subsidiaries more often.
3. Sample selection, data and descriptive statistics
3.1 Sample Selection and Data
SubsidiaryDebt Issues. Using Securities Data Corporation (SDC), I identify all
public and 144a subsidiary bonds and debentures issued between 1985 and 2003. I
exclude issues by special purpose vehicles, branches, trusts and finance subsidiaries,
leaving 3,669 debt issues. I ignore bank loans because machine-readable loan databases
do not indicate whether issuers are subsidiaries. 8
Matching issues to Compustat segments. I need to match subsidiaries with debt
outstanding to divisions of firms, using business segments in the Compustat database to
proxy for divisions. I first attempt the match using the North American Industrial
Classification System (NAICS) codes during the 1990-2003 period. I use NAICS codes
rather than SIC codes because the former are more precise, and NAICS better reflect
economic reality during my sample period than do SIC codes. 9 I begin my segment panel
in 1990 because I need to later merge my data with the Fixed Income Securities Database
(FISD), which only contains debt with maturity dates beginning in 1990. If I cannot
match a subsidiary to a business segment, I attempt to match it to an operating segment. I
match using the highest level of NAICS code precision possible: first I attempt a 5 digit
match, then a 4 digit match, and finally a 3 digit match. If a subsidiary fails to match any
non- financial segments of the parent firm, or if it matches all such segments, I discard it.
Next, I hand check the data using SEC filings to ensure that each subsidiary is associated
with the correct segments for each year.10
Indenture data. In order to get information on bond retirements, guarantees, and
parent company cross-default provisions, using cusip, I merge the above sample with the
8 It is possible that loans, because of concentrated ownership, are less effective at improving internal capital
markets. As Bolton and Scharfstein (1996) argue, with concentrated debt holdings, covenants and other
terms are more easily renegotiated, making them less effective at preventing poaching or underfunding.
9 See Census Bureau (1993) for a discussion of the shortcomings of SIC codes. See Office of Management
and Budget (1994) for a discussion on how the new NAICS system provides an improvement.
10In some instances a subsidiary will get matched to only a subset of the segments it encompasses. In other
instances, segments in a line of businesses similar to the subsidiary are erroneously matched to it. In
addition, some non-subsidiary issues are mislabeled as such. I correct all such errors by hand.
Fixed Income Securities Database (FISD). 11 After the merge, I am left with 1,117 bonds
and debentures.
Assembling the treatmentpanel. Using issue and retirement dates from FISD, I
ensure that a subsidiary bond or debenture is only associated with a given segment in
years that fall between the year the debt was issued and retired, inclusive. If a firm has
subsidiary debt for at least one segment-year observation, then all of its non- financial
segments are in the sample for all the years the firm exists in Compustat, even in years
where no segments have debt. For each segment-year observation in this treatment
sample, I then obtain sales, operating income before depreciation, depreciation, and
capital expenditures, as well as beginning-of-period assets and lagged sales. If the
beginning-of-period assets are unavailable, I estimate them by subtracting capital
expenditures from end-of-period assets and adding depreciation. The results do not
change if I delete these observations. I define sales growth (sg)as the percentage change
in current year sales over the previous year's. I define a segment's return on assets (roa)
as the ratio of operating income before depreciation to beginning-of year-segment assets.
I also obtain the parent's property plant and equipment, credit rating, beginning-of-year
assets, and beginning of year market value of equity. I require each of the above
variables in order to keep a segment-year in my sample. After applying the above sample
criteria, I am left with 2,044 segment-year observations. I label this as my treatment
sample, which includes 121 firms.
The controlpanel. For every firm-year in my treatment sample, I identify the ten
multisegment firms that never had any subsidiary debt, have parent-level public debt, are
I Some issues must be matched by hand because FISD only has the latest cusip of an issue, whereas SDC
has only the historical cusip.
closest in market capitalization to the treatment firm in that year, and are not matched
with another treatment firm-year observation. Of those ten, I keep in my control sample
the two firms closest in tangible assets (PPE/assets) to the treatment firm. I match control
firms with a treatment firm even in years where the latter has no subsidiary debt. I obtain
the same variables for the control sample as I do for the treatment sample. I do not match
using industry because I use industry variables in my tests, and I need variation in these
variables for statistical power. I match on size and tangible assets because both these
variables are related to debt capacity. My results hold if I do not match and instead
include in the control sample all multi-division firms with a market capitalization over $1
billion
Proxiesfor q. As a proxy for a segment's Tobin's q, I use the beginning-of-year
aggregate q of all standalone firms in Compustat that have the same three-digit NAICS
code as the segment. I define q as the ratio of the market value of equity plus book
liabilities to total book assets. The results do not materially change if I define q as the
ratio of the market value of equity to book equity. The results also do not change if I use
the market-capitalization-weighted average q of all firms in the industry, or the median.
Segment volatility. Previous research has used stock return volatility as a proxy
for information asymmetry about risk (Hadlock and James, 2002). I cannot observe
segment volatility, so I use the average enterprise-value-weighted de-levered volatility
(vol) of all standalone firms in a segment's 3-digit NAICS industry. 12 I de- lever because
the effect of leverage on volatility is distinct from the effect of information asymmetry.
To de-lever volatility, I multiply by the ratio of the beginning-of-year market value of equity to the sum
of book debt and market value of equity. My results are not sensitive to de-levering.
12
Segment risk. Credit ratings are not available for all segments, so I use the median
S&P credit rating for all stand-alone firms with the segment's 3-digit NAICS. To proxy
for the degree to which segments in the firm vary in credit risk (divcr), I compute the
absolute difference in industry credit rating grade between the riskiest and safest
segments in the firm To proxy for the degree to which a segment is riskier than the rest
of the firm (crdiff), I compute the difference in credit rating grade between the segment's
industry rating and asset-weighted average for the rest of the firm.
A potential concern is that industry credit rating, in addition to a segment's
business risk, reflects the leverage of a segment's industry. The segment, however, likely
shares with its industry peers those characteristics that determine leverage. These
characteristics likely affect parent leverage, which in turn affects risk-shifting incentives.
Thus industry credit rating is an appropriate risk measure for my purposes.
Appending the control panel to the treatment panel, I am left with a dataset of
6,427 segment-year observations. I define a trinomial categorical variable, wd, to
indicate whether a segment has subsidiary debt and whether it has a parent guarantee.
Following the practice of Standard & Poor's, I treat parent guarantees and
contemporaneous parent public debt with cross-default provisions identically, and
henceforth I refer to them collectively as "guarantees." Parent cross-default provisions
trigger default on parent debt should a subsidiary default, providing an incentive for a
parent to bail out a distressed subsidiary (Samson et al, 2005).
Proxiesfor debt capacity. The standard proxies for debt capacity in the corporate
finance literature typically include size, profitability, investment opportunities, and asset
tangibility. I include proxies for each. For size, I use the market capitalization of the
parent firm, as well as the relative size of the segment, relsize, defined as the ratio of
segment to parent beginning of year assets. For profitability, I include segment roa. I
include the segment's level of sales growth and industry q to proxy for its investment
opportunities. I use the ratio of segment depreciation to beginning of year assets to proxy
for asset tangibility. In addition, a segment may be in an industry where debt is used
sparingly. To control for this effect, I include a dummy, lowlvg, which indicates the
segment is in the lowest leverage quartile industry in a particular year.
Proxiesfor diversity. As proxies for diversity of investment opportunities within
the firm, I compute the intra- firm asset-weighted standard deviation of sales growth and
q. Other studies have used alternate proxies, such as the number of segments or number
of digits by which segment industry codes differ. I do not use such proxies because the
theories relevant to subsidiary debt imply that it is dispersion of investment opportunities
that matters, for which standard deviation in q and sg are the most obvious measures.
The results are qualitatively unchanged if instead I use mean absolute deviation
Other variables. I define also variables qdiff, sgdiff, roadiff and voldiff as the
difference between the segment's q, sg, roa, and vol, respectively, and the asset-weighted
average for all other segments in the firm. All variable definitions are listed in Table 2.
3.2 Descriptive Statistics
Table 3 presents descriptive statistics on firms with and without subsidiary debt.
Firms with subsidiary debt tend to have more tangible assets (fppe) and lower q. Firms
with subsidiary debt also tend to be larger. If the subsidiary debt is not guaranteed, firms
tend to have lower credit ratings, whereas if the debt is guaranteed their credit ratings are
about the same as non-subsidiary-debt-using firms. This result lends preliminary support
to all classical capital structure rationales for subsidiary debt use. Firms with subsidiary
debt tend to be slightly more diverse in sg than those without, lending support to the debt
overhang and internal capital market rationales for subsidiary debt use. The segments of
firms with non- guaranteed subsidiary debt also tend to vary more in their industry credit
risk (divcr), lending support to the risk-shifting rationale for subsidiary debt use.
Table 4 presents descriptive statistics on segments with and without subsidiary
debt. Segments without subsidiary debt tend to have roa no different than other segments
on average, but segments with subsidiary debt tend to have their roa higher by 1 to 2
percentage points, as measured by roadiff Segments with subsidiary debt tend to have
more assets than segments without, and segments without tend to represent a lower
proportion of their parent firm's assets than segments with subsidiary debt. The variable,
relsize, which represents the ratio of segment to firm assets, equals 43% and 29% for
segments with and without subsidiary debt, respectively.
Like firms with subsidiary debt, segments with subsidiary debt tend to have lower
q than segments without. However, segments with guaranteed subsidiary debt tend to
have higher sales growth. In addition, segments with guaranteed subsidiary debt tend to
have higher sales growth relative to other segments within the same firm, as measured by
sgdiff, providing support for the internal capital market rationale for subsidiary debt use.
Recall, that both the RSZ and S&S imply that firms can reduce internal capital market
inefficiencies by placing debt on the investment opportunity-rich subsidiary.
Utility segments appear much more likely to have subsidiary debt. Thirty-six
percent and 45%, respectively, of segments with non- guaranteed and guaranteed debt are
utilities, as opposed to 9% of segments without debt. I therefore include a utility dummy
in my tests to control for possible effects of regulation. However, I keep utilities in my
sample because of their large number and because there is no a priorireason to believe
that the factors influencing non-utilities to have subsidiary debt do not also affect utilities.
Utilities and non-utilities have comparable variation in the variables that I hypothesize to
influence whether a segment has debt. Nevertheless, as a robustness check, I verify that
all my results hold if I exclude utilities from my sample.
4. Empirical tests and results
4.1. Multinomial logistic regression analysis of the determinants of guaranteedand nonguaranteeddebt
In this section, I test my hypotheses about segment and firm characteristics that
determine subsidiary debt use. I use multinomial logistic regression analysis to model the
probability of a segment having guaranteed and non-guaranteed debt. In section 4.1.1 I
describe my methodology in detail. In section 4.1.2 1 describe my results.
4.1.1 Methodology
All my multinomial logistic specifications use a categorical dependent variable, wd, that
can take on one of three values: 0 if a segment has no debt, I if it has non-guaranteed
subsidiary debt, and 2 if it has parent-guaranteed subsidiary debt. The multinomial
logistic regression method models the effect of explanatory variables on the probability
of wd taking on each of three values, requiring that the probabilities sum to 1. All of my
specifications are variations on the following:
P(wd = 1, 2) =
A(a+p 1 qdiff+P2voldiff+p3junk+04divcr+ sdivq+p
5
4roadiff+IrControls
1 )+E(M-Model 1)
Where
Controls1 -::<q, relsize, roa,lowig,vol,lowlvg,dp/assets,norate,fppe,log(size),utility>
T
and
r is a coefficient row vector. All other independent variables are defined in Table 2.
Other specifications include M-Models 2-4, which include all of the explanatory
variables of M-Model 1 as well as certain interaction terms. M-Models 5-8 are identical
to M-Models 1-4, except that they use sales growth (sg) as a proxy for investment
opportunities instead of Tobin's q.
At first glance, it appears inefficient to use as the dependent variable a subsidiary
debt dummy rather than some measure of subsidiary debt outstanding. In treating all debt
levels as equivalent, I ignore a large amount of information. This loss of efficiency,
however, is justified because none of the theories that motivate my rationales for
subsidiary debt have clear implications about how much debt should be placed on a
subsidiary. For instance, if the CEO is using use of proceeds or disclosure requirements
of subsidiary debt to commit to investing in a subsidiary and/or not poaching it, then the
debt level need only be high enough so that debt holders demand such provisions. The
lack of specific predictions about debt level makes it highly likely a model using this
variable will be misspecified. In contrast, the theories do provide concrete predictions
about which segments should have subsidiary debt.
That it is costly to retire public debt before maturity raises another concern:
segments may continue to have subsidiary debt outstanding after its benefits have ceased.
Therefore, at first glance, modeling the probability whether a segment issues debt, rather
than whether it has debt outstanding, seems a superior research design. This approach,
however, has greater drawbacks. First, it ignores the firm's decision to keep debt at a
segment rather than retire it. Twenty-four percent of the issues in my sample were retired
more than a year before maturity. Therefore the costs of early retirement are not so high
as to render unimportant the decision not to retire debt. Second, CEOs are aware of costs
of early debt retirement. Therefore, they likely issue debt at segments expected to persist
in those attributes making subsidiary debt beneficial. Analyzing merely the issuance
decision would thus ignore a large amount of data, reducing power. Nevertheless, as a
robustness check, I re-run my tests using as the dependent variable dummies indicating
whether a segment issues guaranteed or non-guaranteed subsidiary debt in a given year.
My results are unchanged.
To control for time-varying macroeconomic factors, I include year fixed effects in
all specifications. Previous research suggests that such factors affect firm debt policy
(White, 1974 and Henderson, Jegadeesh and Weisbach, 2004). I do not use firm fixed
effects because a large number of dummy variables relative to the total number of
observations can lead to inconsistent estimators in non- linear models (Wooldridge, 2002,
p. 484). Year fixed effects present no problems because they are few in number.
In each specification, I use firm clusters to compute standard errors, making them
robust to intra-firm cross-sectional as well as serial error correlation. My results do not
change if I use year, firm-year, or segment clusters.
4.1.2 MultinomialLogistic Regression Results
4.1.2.1 M-Models 1 and 5
The parameter estimates of M-Model 1 are given in Table 5, Panel A, columns I
and 2. The estimates for M-Model 5 are given in the same columns of Panel B. The
results strongly support the internal capital market rationale for subsidiary debt. They
also provide support for the asymmetric information and risk-shifting rationales for
subsidiary debt use. They provide limited support for the debt overhang rationale. I now
discuss the implications of my results for each of these rationales in turn.
Internal CapitalMarkets. The coefficient on qdiffis highly positive and
significant at the 1% level for guaranteed debt, meaning segments with q much higher
than the rest of their firm are more likely to have subsidiary debt outstanding. Recall that
both the S&S and RSZ theories of internal capital markets imply that firms can improve
internal capital markets by issuing guaranteed subsidiary debt at divisions with the
highest level of q within the firm. The coefficient estimates are also economically
significant. When all variables are at their means, a one standard deviation increase in
qdiffincreases the probability that a segment has guaranteed subsidiary debt by 0.016,
which is large compared to the unconditional probability of 0.044. That qdiff is
significant for non- guaranteed debt runs contrary to the theories, but this finding likely
results from some subsidiary issues having implicit guarantees that I fail to observe. The
results for sgdiff in M-Model 5, presented in Panel B, are qualitatively similar.
Information Asymmetry. The coefficient on voldiff is highly negative and
statistically significant at the 10% level for non-guaranteed debt, meaning that firms are
more likely to use subsidiary debt on a segment if it is in a industry that is much less
volatile than that of other segments in the firm. In so far as volatility proxies for
information asymmetry, this result implies that firms tend to place non-guaranteed debt
on segments with lowest information asymmetry. Also as predicted by the rationale, the
coefficient is not significant for guaranteed debt. The results are also highly
economically significant. When all variables are at their means, a one standard deviation
decrease in voldiff increases a segment's probability of having non- guaranteed debt by
0.37, which is very large compared to the unconditional probability of 0.08. The results
for voldiff in M-Model 5 are qualitatively similar.
Risk Shifting. The coefficient on divcr is positive and significant at the 5% level
for non-guaranteed debt, meaning that firms are more likely to use non-guaranteed
subsidiary debt if their segments are in industries that differ greatly in their median credit
ratings, just as the risk-shifting rationale predicts. That divcr is only significant for nonguaranteed debt indicates that firms only use this sort of debt to prevent risk-shifting.
The economic significance for divcr, however, is only moderate. A one standard
deviation increase in divcr increases the probability that a segment has non-guaranteed
debt by 0.003, compared to an unconditional probability of 0.08.
Debt Overhang. Recall from section 2 that the debt overhang rationale predicts
that firms more diverse in q and sg should use non-guaranteed subsidiary debt. They
should also issue it at segments that have a low q and sg relative to the rest of the firm.
Contrary to this hypothesis, divq is not statistically significant. That the coefficient on
qdiff is significantly positive for non-guaranteed debt further contradicts this rationale.
Both these anomalous results might be due to my inability to detect implicit guarantees.
The coefficient on the junk dummy is positive and highly significant, both
statistically and economically, for non-guaranteed debt. In M-Model 1, having a parent
with a junk credit rating increases the probability of a segment having debt by 0.13. The
result for M-Model 5 is qualitatively similar. This result is consistent with the debt
overhang, risk shifting, and asymmetric information rationales.
4.1.2.2 Other Specifications
In M-Models 2 and 6, I add the variable crdiffto the specification in order to test
whether the relative credit risk of a segment's industry affects its probability of having
subsidiary debt outstanding. Recall, crdiff is the difference in credit rating grade between
the segment's industry median and the asset-weighted average of the industry medians of
all other segments in the firm. A higher value of crdiff indicates the segment has higher
relative risk. In no case is the coefficient on this variable significant. In conjunction with
the significantly positive estimate of the coefficient on divcr, this result implies that firms
tend to place non-guaranteed debt on all segments when segments differ greatly in credit
risk. They do not appear to place guaranteed debt on safer segments and non-guaranteed
debt on riskier segments, but rather place non-guaranteed debt on all.
The results of M-Models 2 and 6 also lend support to the information asymmetry
rationale. While a significantly negative coefficient on voldiff in M-Models 2 and 5
could indicate that lower information asymmetry makes segments more likely to have
subsidiary debt, it might also indicate that the lower relative risk is responsible. That
crdiffis not significant casts doubt on this alternative hypothesis.
In M-Models 3 and 7, I refine my test of the debt overhang rational by adding the
interaction betweenjunk and qdiff and sgdiff. The debt overhang rationale predicts that
the coefficients on these interaction terms should be negative. Placing debt on the low q
or low sales growth subsidiary to avoid debt overhang is going to be a higher priority
when the probability of distress is higher, as indicated by a junk rating. In both models,
however, the coefficients are statistically indistinguishable from zero.
RSZ postulate that segments with many investment opportunities are at higher
risk of poaching when they have a large proportion of the firm's resources. Thus the
RSZ theory implies that firms should use subsidiary debt to protect from poaching those
segments that are both rich in investment opportunities and have a large proportion of the
firm's assets. To test this empirical hypothesis, I estimate M-Models 4 and 8, which are
identical to M-Models 1 and 5, except they include terms that interact qdiff and sgdiff
with relsize, the ratio of segment assets to firm assets. The estimated interaction effect is
positive and significant in the qdiffregression, but only for non-guaranteed contrary to
the prediction of the RSZ model. It is statistically indistinguishable from zero in all other
cases.
To examine whether it is primarily cross-sectional or longitudinal variation
driving my logistic model results, I re-estimate all models that use the time-series
segment means as independent variables. The results, which I do not report, remain
unchanged. I thus conclude that it is primarily cross-sectional variation in segment and
firm characteristics that determines subsidiary debt use.
4.2 Investment-Cashflow Sensitivities
Following the method of Shin and Stulz (1998), I examine the relation between
the presence of subsidiary debt and the sensitivity of the segment investment to other
segment cash flow. If guaranteed subsidiary debt protects the indebted segment from
poaching and/or underfunding, having guaranteed debt outstanding should reduce the
effect of a segment's cash flow on investment in other segments. My findings support
this hypothesis. In section 4.2.1 I describe my tests and findings in detail. In sections
4.2.2 and 4.2.3, respectively, I address omitted variable and endogeneity issues.
4.2.1 The Tests
4.2.1.1 CF Models 1-3 (Table 6)
As a first pass, I estimate the following panel data regressions:
capx = a + fi cf + f 2ocf + fifjwg + controls + E (CF-Model 1)
± fwgxocf + controls + E (CF-Model 2)
capx = a + flcf + f 2ocf + fl3fwg +
Where capx is a segment's capital expenditures in a given year, cf is the segment's
operating income before depreciation, ocf is the total cf for all other segment's in the
firm. All three of these variables are normalized by beginning-of-year segment assets.
The variablefivgd is a dummy indicating that the firm has guaranteed subsidiary debt
outstanding at at least one segment. Controls include q, log(size), andjunk. I include
fixed effects for 3-digit NAICS industry and years, and I cluster standard errors by firm
to ensure they are robust to serial and intra- firm cross-sectional correlation. The results
are robust to firm fixed effects and different standard error clusters, including year, firmyear, and segment.
The results are presented in Table 6. CF-Model 1 measures direct effects.
Consistent with prior literature, segment investment (capx) is positively related to its own
cash flow, q, and the cash flow of other segments, as indicated by the positive
coefficients on cf, q, and ocf respectively. CF-Model 2 includes the interaction between
fwgd and ocf The coefficient on the interaction term is negative. Thus the presence of
guaranteed subsidiary debt is associated with lower capx sensitivity to other segment cash
flows, and hence lower cross-segment capital flows. However, the coefficient is not
statistically significant.
The lack of statistical significance is not surprising, however. If firms are using
guaranteed subsidiary debt to protect certain segments from poaching, one would only
expect to see a reduction in capital transfers out of the indebted segments. Transfers into
the indebted segment, however, should persist under certain circumstances. For example,
if the indebted segment encounters financial difficulties, the guarantee obliges the parent
to transfer capital into it. CF Model 2, however, treats all transfers into and out of a
segment as equivalent, biasing the test against finding an effect.
I partly refine the test with CF Model 3. Instead of interacting ocfwith a dummy
merely indicating the presence of guaranteed subsidiary debt in the firm, I interact it with
a dummy, gcfpos, indicating that the firm has guaranteed subsidiary debt and that the
indebted subsidiary is cash flow positive. A negative coefficient on this term would
indicate that the presence of guaranteed subsidiary debt decreases total cross-segment
capital flows within the firm when the indebted segment is cash flow positive. Consistent
with the hypothesis, I estimate the coefficient on the interaction term to be negative and
economically significant, taking a value of-0.002, which is equal in magnitude to the
estimate of the coefficient on ocf in Model CF 1.
The interaction term in CF Model 3, however, is not statistically significant either.
This result, however, is not surprising, given that the model still imperfectly measures the
extent to which guaranteed subsidiary debt reduces capital transfers out of but not into
and indebted segment. It is easy to envision instances in which a firm would maximize
shareholder value by transferring capital into a subsidiary with guaranteed debt, even
when the subsidiary is not in distress. I thus further refine my tests in the next section.
4.2.1.2 CF Models 6-8 (Table 7)
CF Models 1-3 treat all inter-segment capital transfers equivalently. Guaranteed
subsidiary debt, however, should only influence transfers of capital out of the indebted
subsidiaries. To directly measure this effect, I break down ocf according to segment
status. I define variables cf nodebt, cf_debtnoguar, and cf guaras, respectively, cash
flow of other segments that have no subsidiary debt, that have non-guaranteed subsidiary
debt, and that have guaranteed subsidiary debt. For the sample of segments that belong
to firms with guaranteed subsidiary debt at at least one segment, I run the following panel
data regression:
capx = oa + Plcf+ f 2 cf_nodebt + fl3 cf_debtnoguar+fl3cfguar+controls+E(CF-Model 6)
As before, IIinclude time and industry fixed effects, and I cluster standard errors by firm.
The results are presented in Table 7. The coefficients on cf_nodebt and cf_debtnoguar
are not statistically significant individually, but they are jointly significant at the 5% level
and statistically indistinguishable from one another. In contrast, the coefficient on
cf guar is indistinguishable from zero. Furthermore, the difference between the
coefficient on cf_nodebt and cfguar is positive and statistically significant with a onesided p-value of 0.073. Thus cash flows from segments with guaranteed debt do not get
transferred to other segments, in contrast to cash flows from segments with no debt.
CF Model 6 indicates that capital tends to flow out of segments with nonguaranteed debt just as much as debt-free segments. Thus, consistent with the internal
capital market rationale for subsidiary debt, firms use only guaranteed debt to protect
segment cash flows from diversion to other segments. This inference begs a question:
what mechanism makes guaranteed but not non-guaranteed debt have this effect? It is
not stricter covenants. Examining the public filings of each subsidiary, I find that
covenants on transfers to the parent are on average equally strict for guaranteed and nonguaranteed subsidiary debt. I propose three other possibilities. First, firms that use
guaranteed debt might also tend to use other cash-flow-protecting mechanisms that I fail
to observe. I elaborate on this possibility in the next section. Second, segments with
guaranteed debt might be financed so as to leave them with lower interest coverage.
Finally, the disclosure requirements associated with subsidiary debt might make it more
difficult to allow poaching of high q segmerts. Recall from section 4.1 that segments
with parent- guaranteed subsidiary debt are more likely to have high q, so outside
investors will be more likely to object to the poaching of such segments.
Because cf_nodebt and cf_debtnoguar are jointly significant but statistically
indistinguishable from each other, I pool them into a single variable, cfnoguar and use
this variable in their stead in CF-Model 7. The results are in column 2 of Table 7. As in
CF Model 6, the coefficient on cfguar is statistically indistinguishable from zero.
However, the coefficient on cf noguar is positive and statistically significant at the 5%
level. The difference between the two coefficients is also positive and statistically
significant at the 5% level. Thus, consistent with CF-Model 6, cash flows of segments
without guaranteed debt are significantly related to investment in other segments, but
cash flows of segments with guaranteed debt are not.
Next, I test whether the effect of guaranteed subsidiary debt has an asymmetric
effect on cross-segment capital transfers. I define a dummy variable gcfneg, which
equals 1 if the segment with guaranteed debt has negative cash flows. I then run
regression CF-Model 7, which is identical to CF-Model 6, except that it includes the
interaction between gcfneg and cfguar. As predicted, the coefficient on this interaction
is positive and significant, which implies that funds flow into the segment with
guaranteed debt when it enters distress. The coefficient on cfguar is now negative and
significant, which implies that capital tends to flow into segments with guaranteed debt
even when such segments are performing well. The economic significance of this effect,
however, is small, the coefficient taking a value of only -0.001.
4.2.2 Omitted Variables
The above analysis demonstrates that guaranteed subsidiary debt is associated
with fewer capital flows out of a subsidiary. It is possible, however, that omitted
variables correlated with the presence of guaranteed debt, and not guaranteed debt itself,
are causing this effect. One likely omitted variable candidate is a minority equity stake in
the subsidiary. Minority equity holders might insist on protections that would tend to
prevent poaching of the subsidiary. Debt covenants restricting dividends or distributions
to the parent are another possible omitted variable worth examining.
To see whether these variables drive my results, I examine the public filings of all
firms with subsidiary debt in my sample. I define a dummy variable, meqty, which
indicates whether outsiders hold either a public or private minority stake in a segment. I
also define a dummy variable, div_rst, which indicates whether a segment has debt
covenants that limit dividends or distributions to the parent to 50% of net income or less.
In some cases, covenants completely prohibit dividends.
I then estimate CF Models 4 and 5, which are identical to CF Model 3, except
they exclude firms whose subsidiaries have strict dividend restrictions and minority
equity stakes, respectively. The results are identical to that of CF Model 3.
I estimate CF Models 9 and 10, which are identical to CF Model 7, except,
respectively, they include the interaction of cfguarwith div_rst and meqty. The
coefficient on div rstxcf_guar in CF Model 9 is not significantly different from zero,
implying that dividend restrictions are not driving the results in CF Models 6-8. The
coefficient on meqtyxcf_gaur in CF Model 10 is positive and significant, the opposite of
what it would be if minority equity stakes were driving the results in CF Models 6-8.
Minority equity stakes and dividend restrictions are only two potential omitted
variables. I cannot rule out the hypothesis that firms, when they use guaranteed
subsidiary debt, undertake other measures to protect a subsidiary's cash flow that I fail to
observe and that are not directly tied to the debt. Since a guarantee indicates a firm is
highly committed to the subsidiary, this hypothesis has some plausibility. Even if it were
true, however, my results would still imply that, at the very least, guaranteed subsidiary
debt is often part of a package that protects high-q segment cash flows from diversion.
4.2.3 Endogeneity
The firm's decision to issue parent-guaranteed debt at a segment is endogenous.
As an alternative hypothesis, one could assert that firms issue such debt, for reasons
unrelated to poaching, at a segment when the latter's cash flows are less correlated with
other segments' investment opportunities. Thus indebted segment cash flows might be
less correlated with other segments' investment for a reasons unrelated to the antipoaching effect of subsidiary debt. My results also indicate, however, that the cash flows
of debt- free segments significantly affect the capital expenditures of all other segments,
including segments with parent-guaranteed subsidiary debt. Thus endogenous lack of
correlation of indebted segments cannot explain my results.
It is also possible that firms do not divert capital from segments with guaranteed
debt because it is efficient not to do so. Recall, these segments tend to have the highest q
in the firm. This hypothesis, however, runs counter to Shin and Stulz (1998), who find
that diversified firms do not protect the capital budgets of high q segments.
4.3 SubsidiaryDebt and M&A Transactions Costs
Table 7 compares the acquisition and divestiture activity of firms that do and do
not use subsidiary debt. From Compustat I collect a sample of firms with multiple
segments from 1985-2003. Using SDC, I identify those that did and did not issue at least
one subsidiary debt security over this time. Finally, using SDC, I count the number of
acquisitions and divestitures each firm made since 1990. I find that subsidiary debt users
on average made two more acquisitions and divestitures than non-users. This difference
is statistically significant at the 5%level. Thus firms use subsidiary debt in a manner
consistent with reducing M&A costs or preparing subsidiaries for divestiture.
5. Discussion of alternative rationales for subsidiary debt
FederalTaxes. Subsidiary debt cannot reduce federal tax liability if the parent
files a consolidated tax return or the subsidiary is a non-taxable pass-through entity. A
corporate subsidiary at least 80% owned and domestically incorporated is eligible for tax
consolidation. 13 A subsidiary organized as a partnership, limited liability company
(LLC), or unlimited liability entity is not taxable.14 Using public filings, I find that all
but four subsidiaries in my sample are either non-taxable or eligible for tax consolidation.
State Taxes. Firms operating in multiple states can reduce their state tax liability
by placing debt on subsidiaries in high tax states. However, intra-company debt suffices
for this purpose (Brunori, 2001, p. 113 and Pomp, 1998). Thus state taxes provide no
rationale for the external, public subsidiary debt studied here.
Lower Risk. Subsidiary debt at the safer division has a claim on that division
senior to that of all other securities. Thus such debt sometimes may be safer than parent
13 Scholes, Wolfson, Erickson, Maydew, and Shevlin, p. 297.
14
Ibid, pp. 64 and 80.
debt and thus carry a lower required rate of return. Use of such debt, however, does not
lower a firm's total cost of capital. At best, it transfers risk from subsidiary creditors to
holders of other securities, leaving total risk and cost of capital unchanged. Hence the
low risk of some subsidiary debt per se provides no rationale for its use.
6. Conclusion
This study is the first to examine why firms issue public debt at non- financial
subsidiaries. I find that diversified firms tend to place parent-guaranteed debt on a
segment if it has investment opportunities better than the rest of the firm. I also find that
positive cash flows of segments with guaranteed debt have less influence on investment
in the rest of the firm than do the cash flows of segments without guaranteed debt. Taken
together, these results provide compelling evidence that firms use guaranteed subsidiary
debt, at least in part, to protect their investment opportunity-rich subsidiaries from the
underfunding and poaching problems modeled in Scharfstein and Stein (2000) and Rajan,
Servaes and Zingales (2000).
These results have important implications in the debate over the efficiency of
internal capital markets. Firms would not use subsidiary debt as implied by the theories
of Scharfstein and Stein and Rajan et al. if the postulated internal capital market
inefficiencies were not real. My results, therefore, suggest that such inefficiencies do
exist, but that firms at least partly mitigate them with subsidiary debt.
I also find that diversified firms are more likely to use non-guaranteed subsidiary
debt if they have junk credit ratings and when their segments vary significantly in
Tobin's q and industry credit risk. Taken together, these findings suggest that firms use
subsidiary debt to limit costs associated with debt overhang and risk shifting. Suggesting
that firms use subsidiary debt to limit asymmetric information costs, I find that a segment
is more likely to have subsidiary debt if it operates in an industry much less volatile than
the rest of the firm. Finally, I find that firms that use subsidiary debt tend to more
frequently buy and sell subsidiaries than do firms with parent debt, consistent with the
hypothesis that firms use subsidiary debt to reduce costs of transacting in subsidiaries.
References
Berger, Philip and Eli Ofek. "Diversification's Effect on Firm Value." Journalof
FinancialEconomics 37 (1995): 39-65.
Bolton, Patrick and David S. Scharfstein. "Optimal Debt Structure and the Number of
Creditors." JournalofPoliticalEconomy 104 (1996): pp. 1-25.
Brunori, David. State Tax Policy: a PoliticalPerspective. Washington: The Urban
Institute Press, 2001.
Campa, Jose M. and Simi Kedia. "Explaining the Diversification Discount." Journal of
Finance57 (2002): pp. 1731-1762
Chevalier, Judith. "What do We Know About Cross-Subsidization? Evidence from the
Investment Policies of Merging Firms." Unpublished Working Paper, University
of Chicago, 2000.
Census Bureau. "Conceptual Issues." Economic ClassificationPolicy Committee Issues,
paper 1 (1993).
Clarke, Jonathan E, Edward C. Fee, and Shawn Thomas. "Corporate Diversification and
Asymmetric Information: Evidence from Stock Market Trading Characteristics."
Journalof CorporateFinance 10 (2004): pp. 105-129.
Desai, Mihir, Fritz Foley, and James Hines. "Dividend Policy Inside the Firm." NBER
Working Paper 8698, 2002.
Flannery, Mark J., Joel F. Houston and Subramanyam Venkataraman. "Financing
Multiple Investment Projects." FinancialManagement 22 (1993): pp. 161-172.
Goel, Anand Mohan, Vikram Nanda and M.P. Narayanan. "Career Concerns and
Resource Allocation in Conglomerates." Review ofFinancialStudies 17 (2004):
pp. 99-128.
Graham, John R., Michael L. Lemmon and Jack G. Wolf. "Does Corporate
Diversification Destroy Value?" JournalofFinance 57 (2002): pp. 695-719
Gomes, Joao and Dimitry Livdan. "Optimal Diversification: Reconciling Theory and
Evidence." JournalofFinance 59 (2004): pp. 507-535.
Hadlock, Charles and Christopher James. "Do Banks Provide Financial Slack?" Journal
of Finance 57 (2002): pp. 1383-1419.
Hadlock, Charles, Michael Ryngaert, and Shawn Thomas. "Corporate Structure and
Equity Offerings: Are there Benefits to Diversification?" Journalof Business 74
(2001): pp. 613-635.
Harris, Milton and Artur Raviv. "The Capital Budgeting Process: Incentives and
Information." JournalofFinance 51 (1996): pp. 1139-1174.
Harris, Milton and Artur Arviv. "Capital Budgeting and Delegation." Journalof
FinancialEconomics 59 (1998): pp.259-289.
Henderson, Brian, ANarasimhan Jagadeesh and Michael S. Weisbach. "World Markets
for Raising New Capital." NBER Working Paper 10225, 2004.
Jensen, Michael .C. "Agency Costs of Free Cash Flow, Corporate Finance, and
Takeovers." American Economic Review 76 (1986): pp. 323-329.
Jensen, Michael C. and William H. Meckling. "Theory of the Firm: Managerial Behavior,
Agency, and Ownership Structure." JournalofFinancialEconomics 3 (1976):
pp. 305-360.
John, Teresa. "Optimality of Spin-Offs and Allocation of Debt." Journalof Quantitative
and FinancialAnalysis 28 (1993): pp. 139-160.
John, Teresa and Kose John. "Optimality of Project Financing: Theory and Empirical
Implications in Finance and Accounting." Review of QuantitativeFinanceand
Accounting 1 (1991): pp. 51-74.
Kahn, Charles and Andrew Winton. "Moral Hazard and Optimal Subsidiary Structure for
Financial Institutions." JournalofFinance59 (2004): pp. 2537-2575.
Kensinger, John W. and John D. Martin. "Project Finance: Raising Money the Old
Fashioned Way." JournalofApplied CorporateFinance 1 (1988): pp. 69-81.
Lamont, Owen. '"Cash Flow and Investment: Evidence from Internal Capital Markets"
JournalofFinance 52 (1997): pp. 83-109.
Lamont, Owen and Christopher Polk. "Does Diversification Destroy Value? Evidence
from Industry Shocks." Journalof FinancialEconomics 63 (2002): pp. 51-77.
Lang, Larry and Rene Stulz. "Tobin's q, Corporate Diversification, and Firm
Performance." JournalofPoliticalEconomy 102 (1994): pp. 1248-1280.
Maksimovic, Vojislav and Gordon Phillips. "Do Conglomerate Firms Allocate
Resources Inefficiently across Industries? Theory and Evidence." Journalof
Finance57 (2002): pp. 721-767.
Matsusaka, John G. "Corporate Diversification Value Maximization, and Organizational
Capabilities." JournalofBusiness 74 (2001): pp. 409-431.
Myers, Stewart. "Determinants of Corporate Borrowing." Journal ofFinancial
Economics 5 (1977): pp. 147-175.
Myers, Stewart and Nicholas S. Majluf. "Corporate Financing and Investment Decisions
when Firms have Information that Investors do not have." JournalofFinancial
Economics 13 (1984): pp. 187-221.
Nanda, Vikram. "On the Good News of Equity Carveouts." JournalofFinance 46: pp.
1717-1737.
Office of Management and Budget. "Standard Industrial Classification Replacement."
Economic ClassificationPolicy Committee, 1994.
Ozbas, Oguzhan. "Integration, Organizational Processes, and Allocation of Resources."
JournalofFinancialEconomics 75 (2005): pp. 201-242.
Ozbas, Oguzhan. "Integration and Corporate Investment: Empirical Evidence on Rigid
Capital Budgets." Unpublished Working Paper, USC, 2003.
Pomp, Richard. "The Future of the State Corporate Income Tax: Reflections (and
Confessions) of a Tax Lawyer." In The Future of State Taxation, David Brunori,
ed. Washington: Urban Institute Press (1998), pp. 49-72.
Rajan, Raghuram, Henri Servaes, and Luigi Zingales. "The Cost of Diversity: The
Diversification Discount and Inefficient Investment." JournalofFinance 55
(2000): pp. 35-80.
Samson, Solomon B., Scott Sprinzen, and Emmanuel Dobois-Pelerin. "Corporate
Critera-Parent/Subsidiary Links." Standard& Poor's CorporateRatings
Criteria,2005.
Scharfstein, David. "The Dark Side of Internal Capital Markets II: Evidence from
Diversified Conglomerates." NBER Working Paper 6532, 1998.
Scharfstein, David and Jeremy Stein. "The Dark Side of Internal Capital Markets:
Divisional Rent-Seeking and Inefficient Investment." JournalofFinance55
(2000): pp. 2537-2564.
Schoar, Antoinette. "The Effects of Corporate Diversification on Productivity." Journal
ofFinance 57 (2002): pp. 2379-2403.
Scholes, Wolfson, Erickson, Maydew, and Shevlin. Taxes and Business Strategy: a
PlanningApproach, 2 nd ed. Upper Saddle River, NJ (2002): Prentice Hall, Inc.
Slovin, Myron B and Marie E. Sushka. "The Implications of the Equity Issuance
Decision within a Parent-Subsidiary Structure." JournalofFinance 52 (1997): pp
841-857.
Shin, Hyun-Han and Rene Stulz. "Are Internal Capital Markets Efficient?" Quarterly
JournalofEconomics 113 (1998): pp. 531-552.
Stumpp, Pamela, William Rottino, David Fanger, Richard Stephan, and Donald Carter.
"Rating Non-Guaranteed Subsidiaries: Credit Considerations in Assigning
Subsidiary Ratings in the Absence of Legally Binding Parent Support." Moody 's
Rating Methodology (2003): pp. 1-12.
Villalonga, Belen. "Diversification Discount or Premium? New Evidence from the
Business Information Tracking Series." Journalof Finance 59 (2004): 479-506.
Thomson, Robert. "Piercing the Corporate Veil: an Empirical Study." CornellLaw
Review 76 (1991): pp. 1036-1065.
Thomas, Shawn. "Firm Diversification and Asymmetric Information: Evidence form
Analysts' Forecasts and Earnings Announcements." JournalofFinancial
Economics 64 (2002): pp. 373-396.
Vijh, Amand. "The Positive Announcement Returns of Equity Carveouts." Journalof
Business 75 (2002): pp. 153-190.
Vijh, Amand. "Does a Parent-Subsidiary Structure Enhance Financing Flexibility?"
JournalofFinance (2005): forthcoming.
White, William. "Debt Management and the Form of Business Financing." Journalof
Finance 29 (1974): pp. 1467-1484.
Whited, Toni. "Is it Inefficient Investment that Causes the Diversification Discount?"
JournalofFinance56 (2001): pp 1667-1691.
Wulf, Julie. "Influence and Inefficiency in the Internal Capital Market: Theory and
Evidence." Unpublished Working Paper, University of Pennsylvania, 2002.
Table 1
Empirical predictions about characteristics of subsidiary debt issues, issuers, and parents
based on the various proposed rationales for subsidiary debt use. Rationales are listed in
the top row, and prediction categories are listed in the left-most column.
Classical Capital Structure Rationales
Improving Avoiding
Reducing
Lowering
Internal
Debt
RiskAsymmetric
Capital
Overhang
Shifting
Information
Markets
Costs
Never on
riskier
Is Subsidiary Debt
subsidiaries,
Yes
Never
Never
Guaranteed?
Sometimes
on safer
ones.
Quality of Issuer's
Investment
No
No
Opportunities
Good
Poor
prediction
Prediction
Relative to Rest of
Firm
Sensitivity of
segment investment
Lower
to the cash flows of
No
No
No
than
Prediction Prediction
Prediction
other segments that
normal
have guaranteed
subsidiary debt
Diversity in
investment
No
No
opportunities of
High
High
Prediction
Prediction
firms that Issue
Subsidiary Debt
Diversity of credit
No
No
No
risk of firms that
ist
issue subsidiary debt
Credit rating of
issuer's parent
Issuer's Volatility
Relative to Rest of
Firm
Prediction Prediction
No
Prediction
Low
Prediction Prediction
High
Low
Prediction
Prediction
Low
Low
Table 2
Variable Definitions
Beginning of period segment assets; firm-level assets are sum of segment
assetsassets.
crdiff
Difference in credit rating grades between the segment's industry credit
rating and the average for all other segments in the firm. crdiff increases in
a segment's relative risk.
capx
divcr
divq
dp/assets
fppe/assets
junk
lowig
lowlvg
norate
Segment capital expenditures
Diversity of credit risk: absolute difference in industry credit rating grades
between the segment in the firm in the safest industry and the segment in
the riskiest industry. An industry credit rating is the median S&P rating of
all standalone firms with the same 3-digit NAICS code.
Diversity of q, defined as the standard deviation of q of a firm's segments,
divided by a firm q.
Diversity of sales growth, defined as the standard deviation of sales growth
of a firm's segments.
Segment depreciation divided by assets.
Firm property, plant & equipment divided by beginning of year firm assets
Dummy indicating that a firm's S&P debt rating is lower than BBBDummy indicating a low investment grade rating, from BBB+ to BBBDummy indicating that a segment belongs to a 3-digit NAICS industry that
is in the lowest leverage quartile of industries in a given year.
Dummy indicating the firm has no credit rating.
Asset-weighted average of (market equity + book liabilities)/(book assets)
q
roa
roadiff
sg
sgdiff
size
utility
vol
voldiff
for the previous period of all standalone firms with the same 3-digit
NAICS code as a business segment.
Difference between a segment's q and the asset-weighted average q for all
other segments in the same firm for a particular year.
Relative size of a segment, defined as the ratio of a segment's assets to the
firm's.
Segment return on assets, defined as segment operating income before
depreciation divided by segment assets. Firm- level roa is the assetweighted average of segment roa.
Difference between segment roa and the asset-weighted average roa for all
other segments in the same firm for a particular year.
Percentage change in segment sales in the current period relative to the
previous period. Firm-level sg is the asset-weighted average of segment
sg.
Difference between segment sg and the asset-weighted average sg for all
other segments in the same firm for a particular year.
Natural logarithm of a firm's beginning-of-year market capitalization.
Dummy indicating that a segment's primary 3 digit NAICS code is 221.
Enterprise-value-weighted-average de-levered stock price volatility in the
previous year of all standalone firms in the same 3-digit NAICS industry.
Difference between segment vol and the asset-weighted average vol for all
other segments in the same firm for a particular year
Table 3
Descriptive statistics on firms. Sample period is 1990-2003. Panel A includes
firms with no subsidiary debt, and Panels B and C include firms in which
subsidiary debt is not guaranteed and guaranteed by the parent, respectively.
fppe/assets
log(size)
junk
leverage
assets
firm q
firm sg
firm roa
divq
divsg
divcr
#segments
N
1427
1427
1427
1427
1427
1427
1427
1427
1427
1427
1426
1427
Panel A: Firms Without Subsidiary Debt
Mean
Std Dev
Min
25th Pctl 50th Pctl 75th Pctl
0.44
0.222
0.01
0.25
0.39
0.62
7.53
1.728
-4.12
6.77
7.68
8.59
0
0
0
0
0.17
0.380
0.35
0.170
0.00
0.25
0.33
0.41
4908
7947
13
1113
2462
5772
1.73
0.654
0.94
1.28
1.52
1.96
0.344
0.05
0.13
0.04
-0.74
-0.04
0.16
0.22
0.15
0.167
-0.52
0.12
0.13
0.127
0.00
0.01
0.10
0.19
0.16
0.263
0.00
0.03
0.08
0.19
1.10
1.298
0.00
0.00
0.75
1.66
3.20
1.252
2
2
3
4
Max
0.91
12.04
1
1.89
106340
4.48
3.39
0.86
0.75
2.24
9.52
fppe/assets
log(size)
junk
assets
leverage
firm q
firm sg
firm roa
divq
divsg
divcr
#segments
N
382
382
382
382
382
382
382
382
382
382
382
382
Panel B: Firms with non-Guaranteed Subsidiary Debt
Mean
Std Dev
Min
25th Pctl 50th Pctl 75th Pctl
0.55
0.202
0.04
0.42
0.59
0.70
1.794
2.81
6.73
7.90
9.08
7.89
0.16
0.371
0
0
0
0
10439
14506
36
1774
4463
13140
0.38
0.139
0.04
0.30
0.36
0.45
1.50
0.491
0.98
1.19
1.36
1.63
0.05
0.354
-0.75
-0.02
0.05
0.12
0.15
0.069
-0.10
0.11
0.14
0.18
0.112
0.00
0.05
0.12
0.20
0.13
0.314
0.00
0.03
0.09
0.22
0.19
1.23
1.175
0.00
0.40
0.91
1.75
3.54
1.425
3
3
4
Max
0.92
11.77
1
91130
0.90
4.00
2.43
0.42
0.60
2.42
5.92
fppe/assets
log(size)
junk
assets
leverage
firm q
firm sg
firm roa
divq
divsg
divcr
#segments
N
170
170
170
170
170
170
170
170
170
170
170
170
Panel C: Firms with
Mean
Std Dev
0.55
0.182
7.99
1.460
0.13
0.338
10216
16979
0.39
0.153
1.40
0.369
0.07
0.313
0.14
0.064
0.11
0.101
0.23
0.299
1.02
1.195
1.518
3.78
Ill
Guaranteed Subsidiary Debt
Min
25th Pctl 50th Pctl 75th Pctl
0.09
0.40
0.59
0.70
-2.53
7.24
7.86
8.57
0
0
0
0
827
2701
4302
9642
0.11
0.32
0.37
0.44
1.01
1.16
1.24
1.58
-0.75
-0.01
0.04
0.16
0.01
0.10
0.13
0.16
0.00
0.01
0.10
0.17
0.00
0.05
0.13
0.27
1.34
0.00
0.00
0.69
3
3
5
Max
0.85
11.86
1
117384
1.74
3.45
1.37
0.49
0.48
1.73
6.12
Table 4
Descriptive statistics on segments. The sample period is 1990-2003. Panel A
includes segments with no subsidiary debt, and Panels B and C include segments in
which subsidiary debt is not guaranteed and guaranteed by the parent, respectively.
q
sg
roa
vol
crdiff
sgdiff
qdiff
roadiff
voldiff
assets
relsize
dp/assets
lowlvg
utility
N
5757
5757
5757
5757
5687
5757
5757
5757
5757
5757
5757
5757
5757
5757
q
sg
roa
vol
crdiff
sgdiff
qdiff
roadiff
voldiff
assets
relsize
dp/assets
lowlvg
utility
N
519
519
519
519
518
519
519
519
519
519
519
519
519
519
q
sg
roa
vol
crdiff
sgdiff
qdiff
roadiff
voldiff
assets
relsize
dp/assets
lowlvg
utility
N
252
252
252
252
252
252
252
252
252
252
252
252
252
252
Panel A: Segments Without Subsidiary Debt
Mean
Std Dev
Min 25th Pctl 50th Pctl 75th Pctl
Max
1.73
0.724
0.94
1.23
1.52
1.98
4.48
0.08
0.666 -0.74
-0.08
0.04
0.15
4.57
0.14
0.215 -0.52
0.07
0.14
0.23
0.86
0.10
0.040
0.04
0.08
0.09
0.12
0.24
0.31
3.266 -21.00
-1.00
0.00
1.50
21.00
0.02
0.71 -4.69
-0.15
0.00
0.12
5.28
0.04
0.65 -3.35
-0.18
0.00
0.26
3.35
-0.01
0.187 -0.99
-0.09
0.00
0.07
0.86
0.00
0.031 -0.15
-0.01
0.00
0.02
0.19
1639
3109
6
202
624
1641
24555
0.28
0.249
0.00
0.09
0.20
0.41
0.99
0.06
0.043
0.00
0.03
0.05
0.07
0.27
0.15
0.361
0.00
0.00
0.00
0.00
1.00
0.11
0.307
0
0.00
0.00
0.00
1.00
Panel B: Segments With Non-Guaranteed Subsidiary Debt
Mean
Std Dev Minimun 25th Pctl 50th Pctl 75th Pctl Maximum
1.43
0.502
0.95
1.14
1.23
1.55
4.65
0.04
0.422 -0.75
-0.03
0.04
0.12
4.99
0.16
0.135 -0.20
0.10
0.14
0.19
1.53
0.09
0.043
0.04
0.06
0.08
0.11
0.27
-0.55
2.97 -17.00
-2.00
-0.32
0.05
16.37
-0.06
0.57 -4.27
-0.14
-0.01
0.10
5.02
-0.09
0.547 -3.29
-0.28
-0.05
0.03
3.40
0.02
0.177 -0.72
-0.05
0.01
0.08
1.67
-0.01
0.03 -0.13
-0.03
0.00
0.00
0.19
4417
5989
8
538
1728
5656
24583
0.46
0.288
0.01
0.21
0.42
0.71
0.98
0.06
0.039
0.00
0.04
0.05
0.06
0.36
0.04
0.197
0.00
0.00
0.00
0.00
1.00
0.36
0.480
0
0.00
0.00
1.00
1.00
Panel C: Segments with Guaranteed Subsidiary Debt
Mean
Std Dev Minimun 25th Pctl 50th Pctl 75th Pctl Maximum
1.39
0.465
0.95
1.14
1.23
1.54
4.94
0.11
0.730 -0.75
-0.07
0.04
0.16
4.99
0.12
0.207 -0.72
0.07
0.13
0.17
0.89
0.09
0.044
0.04
0.06
0.08
0.11
0.27
-0.24
2.57 -10.00
-1.34
0.00
0.12
15.84
0.05
0.70 -2.12
-0.17
0.00
0.17
5.24
-0.03
0.454 -1.29
-0.19
0.00
0.02
3.06
0.00
0.238 -1.09
-0.06
0.00
0.07
0.91
-0.01
0.03 -0.17
-0.02
0.00
0.00
0.17
3806
5579
27
717
2016
4650
24583
0.43
0.312
0.00
0.12
0.41
0.69
1.00
0.05
0.050
0.00
0.03
0.04
0.06
0.36
0.02
0.125
0.00
0.00
0.00
0.00
1.00
0.45
0.499
0
0.00
0.00
1.00
1.00
Table 5
Multinomial logistic regression analysis on the determinants of whether a segment has
guaranteed or non-guaranteed debt outstanding in a given year. Parameter estimates are
presented in separate panels for regressions that use q and sales growth (sg) as proxies for
investment opportunities. Standard errors are clustered by firm. All regressions include year
fixed effects. Variable definitions are in Table 1.
Table 5, Cont
Panel A
M-Model 1
Non.Guaranteed
Test Variables
qdiff
0.470**
(0.203)
voldiff -6.914*
(3.553)
junk
1.918***
(0.542)
junkxqdiff
M-Model 2
M-Model 3
NonNonGuaranteed Guaranteed Guaranteed Guaranteed Guaranteed
0.996***
(0.341)
-3.859
(4.843)
0.647
(0.815)
0.453**
(0.211)
-6.054
(3.921)
1.896***
(0.543)
0.993***
(0.342)
-3.599
(5.005)
0.636
(0.816)
0.490**
(0.226)
-7.003*
(3.616)
1.911"***
(0.541)
-0.095
(0.319)
0.919***
(0.329)
-3.644
(4.859)
0.620
(0.810)
0.508
(0.445)
relsizexqdiff
divq
1.403
(0.990)
0.057**
(0.024)
0.622
(1.573)
0.026
(0.039)
roadiff 0.470
(0.686)
0.505
(0.777)
divcr
crdiff
Control Variables
relsize
2.327***
(0.294)
-0.792***
q
(0.251)
roa
0.866*
(0.468)
lowlvg -0.833*
(0.448)
dp/asset -0.564:
vol
...
(0.452)
-1.172***
(0.366)
-0.048
(0.573)
-.703**
(2.944)
1.080
-2.474
(3.653)
.(7.071)
0:1974*"
(0.460)
norate 2.608***
(0.500)
fppe/assc :1420"**
(0:.641)
0.403
(0:682)
(0.100)
0.907***
0.518
(0.866)
-0.095
(0.888)
:0.332**
(0.132)
1:.359**
(0.350)
(0.538)
log(size) 0.421 ***
:
2.075***
(0.405)
0.870
(4.148)
lowig
utility
.
.
M-Model 4
NonGuaranteed Guaranteed
-0.035
(0.267)
-6.753*
(3.611)
1.940***
(0.544)
0.837*
(0.466)
-3.839
(4.847)
0.650
(0.812)
(0.024)
0.396
(0.996)
0.675
(1.594)
0.025
(0.039)
0.507
(0.782)
0.577
(0.710)
0.546
(0.807)
2.048***
2.324**
:0.457)
(:i
(0.293).
-:1.207**.*. .-0'.794***
2.069***:
2.425***
(0.445)
(0.290) :
-0.758***
2.095***
(0.460)
-1.159***
(0.367)
-0.045
(0.578)
(0.366).
(0.250)
(0.360)
-1.677**:
(0.250)
0.872*. :
(0.466)
-0.832*"
(0.404):
(0.448):
(0.408)-
0.754 :
-0.570
0.903
(4.063)
(4.146)
-3.154
(2.947) :.
1.113
(7.151)
(3.674)
0.983"*
(0.459)
0.410
(7.099)
0.402
2.619***
0.524
0.975"*
(0.460)
2.609***
0.781"
(0.463)
-0.759*
(0.441):
-0.553
(2.958)
1.067
(3.625)
0.984**
(0.461)
2.61:7***
(0.500)
1.591**
(0.659)
0.422***
(0.098)
0.880**
(0.351)
-0.074
(0.569)
-1.689***
(0.413)
0.839
(4.185).
-2.442
(7.077)
0.404
(0.681)
0.517
1.461
(0.994)
0.052**
(0.026)
-0.025
(0.032)
0.459
(0.692)
0.689
(1.584)
0.023
(0.043)
-0.010
(0.055)
0.510
(0.791)
2.291:***
.:. I
:
•
(0.290)
-0.821***
(0.257)
0.876*
(0.473)
-0.810*
(0.451).
-0.683
(2.983)
0.422
(3.807)
(0.499)
1.439"*
(0.639)
0.419***
(0.100).
0.838**
(q.361)
(0.682)
(0.866)
-0.057
(0.886)
0.329**"
(0.132)
1.288**
(0.527).
1.401
(0.990)
0.057**
(0.024)
0.548
(1.556)
0.026
(0.039)
0.469
(0.686)
(0.501)
1.420'**
(0.640)...
-1.184***
-0.059.
(0.572)
-1.696"**
-2.694
(0.682)
0.519
(0.866)
-0. 102
:.(0.889)
0.421***
0.334**..
(0.100)
(0.132)1.341**
0.910***
(0.351)
Observa
6427
6363
Pseudo-.
0.19
0.19
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
(0.538)
6427
0.19
1.066**
(0.425)
1.544
(0.976)
0.054**
(0.865)
-0.043
(0.885)
0.331"*
(0.132)
1.353**
(0.540)
6427
0.19
Table 5, Cont.
Panel B
M-Model 5
NonGuaranteed Guaranteed
Test Variables
sgdiff
0.111
0.502***
(0.142)
(0.173)
voldif -6.731"
-2.941
(3.532)
(4.832)
junk
1.942"**
0.668
(0.559)
(0.830)
junkxsgrowthdiff
M-Model 6
Nnn-
N on-
Guaranteed Guaranteed
0.104
(0.143)
-6.153
(3.764)
1.924***
(0.561)
0.505***
(0.175)
-2.886
(5.056)
0.661
(0.830)
relsizexsgrowthdiff
uIvsg
s
divcr
sgrow
roa
lowig
fp]pe
size
utility
Pbser'
Pseud(
(0.271)
-6.701 *
(4.814)
0.659
(3.523)
1.953"**
(0.560)
0.023
(0.832)
0.081
(0.560)
(0.327)
-2.951
(4.848)
0.670
(0.831)
(0.146)
(0.178)
(0.025)
0.769**
(0.374)
0.027
(0.042)
-0.439
(0.319)
0.339
(0.355)
0.071"***
(0.024)
-0.087
(0.408)
0.792*
(0.407)
0.028
(0.042)
0.545
(0.673)
0.414
(0.737)
0.502
(0.665)
0.396
(0.733)
2.015***
0.538
(0.676)
2.194***
(0.282)
-0.350*
(0.196)
2.015***
2.163***
1.993***
(0,447)
-0.586**
(0.252)
-0.009
(0.559)
-1.989***
(0.490)
0,624
2.194***
(0.277)
(0.448)
(0.282)
-0.349*
2.011"***
(0.448)
-0.585**
2.188***
(0.283)
(0.196)
(0.253)
0,767*
(0.247)
-0.006
(0.445)
(0.559)
-1.116**
(0.436)
-0.385
(3.007)
1.866
(3.303)
1.010**
(0.462)
2.596***
-1.987"***
(0.490)
0.796*
(0.445)
-1.112"*
(0.436)
-0.394
(3.006)
1.871
(3.290)
1.010**
(0.462)
2.609***
0.768*
-1.116**
-0.384
1.868
(3.304)
1.011"*
(0.462)
norate
0.545*
0.404
(0.731)
(3.007)
vol
0.392
0.542
(0.671)
(0.436)
dp/ass,
0.485***
(0.167)
-2.902
0.779**
(0.372)
0.026
(0.045)
-0.005
(0.059)
0.403
(0.742)
(0.445)
lowly:
0.105
(0.156)
-6.723*
(3.535)
1.942***
0.774**
(0.373)
0.027
(0.042)
0.250
(0.332)
0.068***
(0.025)
-0.020
(0.030)
Controls
relsize
Guaranteed Guaranteed
M-Model 8
NonGuaranteed Guaranteed
0.253
(0.331)
0.071***
(0.025)
crdiff
roadifl
M-Model 7
2,597***
(0.505)
1.697***
(0.610)
0.398***
(0.101)
1.174**
(0.332)
6427
0.18
-0.350*
(0.198)
0.773*
-0,595**
(0.254)
-0,008
(0.449)
(0.560)
-1.118"*
-1.984***
(0.485)
0.568
(0.438)
-0.456
(4.272)
-1.277
(3.035)
1.577
(6.587)
0.479
(0.704)
(3.361)
1.020**
(0.461)
2.606***
(0.503)
1.720***
(0.611)
0.397***
(0.101)
0.525
(0.900)
0.417
(0.927)
0.289**
(0.131)
1.743***
(0.514)
(4.168)
-1.666
(6.654)
0.488
(0.703)
0.530
(0.901)
0.454
(0.928)
0.254
(0.331)
0.071"**
(0.504)
0.594
(4.284)
-1.30 1
(6.566)
0.479
(0.703)
0.524
-0.513**
1.697***
(0.610)
0.286**
(0.900)
0.417
(0.927)
0.398***
0.288**
(0.131)
1.704"**
(0.101)
(0.131)
1.174***
1.704"**
(0.609)
0.403***
(0.102)
1.741"***
1.179***
1.137***
(0,336•r
to if),"
014 (0.336
,U' -Jv ....
6363
0.18
Robust standard errors in parentheses
* significant at 10%; **significant at 5%; ***significant
at 1%
6427 (U.•
0.18
1 5)
(0.446)
-0.614*
(0.319)
-0.003
(0.557)
-1.986***
(0.487)
0.626
(4.267)
-1.274
(6.580)
0.479
(0.703)
0.526
(0.504)
(0.902)
(0.925)
(0.332)
0.417
0.290**
(0.131)
1.743-**
6427
0.18
(0.515)
Table 6
Panel data regression analysis wherein segment capital expenditures, normalized by
assets, are the dependent variable. The segment-year is the unit of observation.
Independent variables include a dummy indicating that the firm has guaranteed debt on at
least one segment (fwg), a dummy indicating that the firm has segments with guaranteed
debt and positive cashflow (gcfpos), the segment's cash flow (cf), and the cashflow of all
other segments within the firm (ocf). Both cf and ocf are normalized by segment assets.
Control variables include segment q, the log of the parent's beginning-of-year market
capitalization, and a junk dummy. All specifications fixed effects for both both calendar
year and 3 digit NAICS codes. Standard errors are clustered by firm.
CF Model 1 CF Model 2 CF Model 3 CF Model 4' CF Model S5
Cashflow Variables
0.003
0.003
0.003
ocf
0.002
0.003
(0.002)
(0.002)
(0.002)
(0.002)
(0.001)
-0.007
-0.011
fwgd
(0.008)
(0.009)
-0.001
fwgxocf
(0.002)
-0.007
-0.007
-0.006
gcfpos
(0.009)
(0.009)
(0.008)
-0.002
-0.002
-0.002
gcfposxocf
(0.002)
(0.002)
(0.002)
0.059***
0.059***
0.057***
0.062***
0.060***
cf
(0.016)
(0.016)
(0.016)
(0.016)
(0.016)
Conirol Variables
0.006
.007
0.007 :
.:0.008
0.007
q
(0.005)
(0.005)
(0.005)
(0.006)
(0.005)
0.000
-0.000
0.000
0.000
0.000
size
(0(0002)
(0.002)
(0.002)
(0.002)
(0.002)
0.013
0.015
0.012
0.012
0.011
junk
(0.010)
(0.011)
(0.010)
(0.010)
(0.0.10)
6031
6156
6347
6347
6347
Observations
0.16
0.16
0.15
0.15
0.15
R-squared
* significant at 10%; ** significant at 5%; *** significant at 1%
'Excludes firms whose subsidiaries have severe dividend restrictions
"Excludes firms that have sold minority equity stakes in their subsidiaries
Table 7
Panel data regression of a segment's capital expenditures (capx) on its own cash flow (cf)
and the cash flows of other segments, where EBITDA proxies for cash flow. Capx and
all cash flow variables are normalized by segment assets. Other segment cash flows are
broken down according to whether other segments have guaranteed debt (cf _guar) or not
(cf noguar). Cash flows of segments without guaranteed debt are broken down into
segments without any debt (cf nodebt), and segments with non-guaranteed debt
(cf_debtnoguar). The dummy variables divrst, meqty, and gcfneg, respectively, indicate
whether the subsidiary with guaranteed debt is restricted in its ability to pay dividends,
whether outsiders hold a minority equity stake in it, or whether it has negative cashflow.
Other variables are defined in Table 2. Regressions include fixed effects for calendar
year as well as 3 digit segment NAICS codes. Standard errors are clustered by firm.
Only firms that utilize guaranteed subsidiary debt are included in the sample.
Panel A: Parameter Estimates & Standard Errors
CF Model 6 CF Model 7 CF Modle 8 Cf Model 9 Cf Model 10
Cashflow Variables
cf nodebt
Exp. Simn
+
0.008
cf_debtnoguar
+
(0.005)
0.006
(0.004)
cfnoguar
+
cfguar
0
0.007**
(0.003)
-0.001
-0.001
(0.001)
(0.001)
gcfneg
gcfnegxcf__guar
0.010***
(0.003)
-0.001***
0.007**
0.007**
(0.003)
(0.003)
-0.001
-0.001*
(0.001)
-0.028*
(0.016)
0.415*
(0.002)
(0.001)
(0.230)
div rst
?
0.002
(0.017)
0.000
(0.002)
div_rstxcfguar
+
0.026
(0.049)
0.027
(0.048)
0.032
(0.048)
0.028
(0.049)
-0.018
(0.018)
0.019"***
(0.007)
0.033
(0.048)
+
0.071**
0.071 **
0.071**
0.071**
0.071**
(0.034)
(0.033)
-0.014
(0.033)
-0.017
(0.019)
(0.034)
(0.034)
-0.016
meqty
meqtyxcfguar
cf
Control Variables
q
junk
?
size
?
-0.015
(0.019)
-0.001
(0.003)
602
Observations
0.32
R-squared
Robust standard errors in parenthases
(0.019)
-0.001
(0.003)
602
0.32
-0.013
-0.002
(0.017)
-0.001
(0.003)
(0.003)
602
0.34
602
0.32
* significant at 10%; ** significant at 5%; *** significant at 1%
(0.018)
-0.001
(0.003)
602
0.33
Table 7, Cont.
Panel B: Tests for Positive Differences Between Parameters
Exp. Sign CF Model 6 CF Model 7 CF Modle 8 Cf Model 9
cfnodebt - cfnoguar - cfnoguar -
Difference Tested
Est. difference
Pvalue (One-sided)
+
cf_guar
0.009
0.073
cf_guar
0.007
0.022
cfguar
0.010
<0. 001
Cf Model 10
cfnoguar -
cfnoguar -
cfguar
0.008
0.043
cf_guar
0.008
0.021
Table 8
M&A activity of Firms that do and do not Use Subsidiary Debt
Use Subsidary Debt?
Mean # of Divestitures
Yes
6.38
(.313)
No
4.37
(.379)
Difference
2.00***
(.491)
Mean # of Acquistions
10.48
(.548)
8.54
(.54)
1.94**
(.769)
s
To be considered a user of subsidiary debt, a firm must have at least one
subsidiary issue debt issue. The number of acquisitions and divestitures
are counted since 1990. Standard errors in parenthesis. *** and **
indicate significance at the 1%and 5% level, respectively.
Chapter 2
Is the Chinese Wall too High? Investigating the Costs of New
Restrictions on Cooperation Between Analysts and Investment
Bankers
1. Introduction
In the summer of 2002, near the end of a two-year long investigation by New
York State Attorney General Elliot Spitzer, a series of new regulations were introduced
that further restrict cooperation and interaction between analysts and investment bankers.
The new regulations are thus said to have raised the "Chinese Wall," the industry name
for the institutional separation of research and investment banking within firms that
provide both services. Recent evidence suggests that the Chinese Wall regulations have
benefits"1 ; they appear to have the reduced the bias of bank-affiliated analysts
documented in the pre-regulatory period. 16 Some policy makers, however, including
Elliot Spitzer, hypothesize that the new regulations may have caused a reduction in
analyst coverage (Parker, 2005). This study tests this hypothesis.
In particular, I examine whether the increase in analyst coverage enjoyed by firms
after an SEO declined during the post regulatory period. Using regression analysis that
controls for various factors, I find that SEO's in the post-regulatory period enjoy just as
large an increase in analyst coverage as in the pre-regulatory period. In addition, I
compare the post-regulatory change in research coverage of recent IPO's, firms that pay
high investment banking fees, to that of a matched-pair control sample of firms that pay
no fees within a four year window around the treatment firm's IPO date. I find that
15 See Kadan, Madureira, Wang, and Zach, 2005 and Yonca, Sunder and Sunder 2005
See Dugar and Nathan (1995), Lin and McNichols (1998), Michaely and Womack (2000), Dechow
Hutton and Sloan (2000), Iskoz (2003), Malmendier and Shanthikumar, (2004), Barber, Lehavey, and
Trueman (2005), Ljundquist, Martson, and Wilhelm (2005), Kolasinski and Kothari (2006).
16
coverage declines by the same amount for the two groups with the adoption of new
regulations. Making the identifying assumption that any adverse effect of a higher
Chinese Wall on coverage should be worse for IPO firms, because they pay high
investment banking fees, I conclude from my results that the higher Wall has no such
adverse effect.
My findings have policy implications. The benefits of analyst coverage on
liquidity and cost of capital are well-documented in the literature (Brennan and
Subrahmanyam, 1995, Yohn, 1998, Roulstone, 2003). Thus it should come as no
surprise that policy makers have expressed concern (Parker, 2005) over the documented
declined in research coverage of recent years (Boni, 2005). There are also strong a priori
to reasons, discussed below, to believe that the new Chinese Wall regulations might
reduce coverage, so it is important to empirically test whether they have this effect. My
findings suggest that policy makers need to look at other factors, including regulations
adopted in recent years unrelated to the Chinese Wall, as the culprits behind the recent
drop in analyst coverage.
Theory suggests the higher Chinese Wall may have reduced analyst coverage by
shutting down a channel by which some firms purchase coverage. Small, illiquid firms
generate little trading revenue and investor demand for research coverage, making it
difficult for them to attract it. Before the new regulations, evidence suggests that
investment banks regularly provided a channel for such firms to purchase analyst
coverage by bundling it with equity underwriting services. In a legal brief, the SEC
asserts, "As part of the package of services it offered to issuers to win investment banking
business...Morgan Stanley typically committed that its analysts would initiate (or
continue) research coverage of the issuer.""17 Krigman, Shaw and Womack (2001)
provide evidence that a commitment to accurate and timely analyst coverage is the most
important factor driving firms' choice of equity underwriter. The new NASD and NYSE
rules 2711 and 472 adopted on July 29, 2002, however, prohibit analyst participation in
equity underwriting deals and prohibit the compensation of analysts out of underwriting
fees. Spizter's legal settlement with the ten largest investment banks included these and
other regulations. 18 Restricting the ability for bankers to pay analysts out of deal revenue,
as well as cooperate with them in doing deals, may have reduced their ability to bundle
underwriting services with analyst coverage. Thus the higher Chinese Wall plausibly
could have reduced the ability of firms that generate little brokerage revenue but high
investment banking fees to attract analyst coverage.
Several new regulations unrelated to the Chinese Wall were adopted around the
same time as those that raised the Wall. By demonstrating that the higher Chinese Wall
likely had little effect, my results suggest these other regulations may be important
factors driving the post-regulatory decline in affected analyst coverage. They include
Regulation FD, adopted in 2000, which requires firms to disclose to the public any
information that they disclose to analysts. Gomes, Gorton, and Maduieira (2004) find
evidence that this regulation reduced analyst coverage for small stocks by lowering
investor demand for such coverage. The Spitzer Settlement and NASD 2711 and NYSE
472, in addition to the new Chinese Wall regulations, contain provisions that make all
analyst coverage more costly. The Spitzer Settlement requires research departments in
17 SEC v. Morgan Stanley, complaint #18117, United States District Court, Southern District, para. 17.
18 See NASD rule 2711 and NYSE rule 472. For a summary of the rules and a history of their adoption,
along with a comparison of their provisions to that of the Spitzer Global Settlement, see the 2005 Joint
Report by NASD and the NYSE on the Operationand Effectiveness ofResearch Analyst Conflict ofInterest
Rules.
banks to have their own, dedicated legal and compliance staff. It also requires that an
oversight committee review all recommendation changes. Finally, it requires that
investment banks, for all firms they cover, provide independent research for their
brokerage clients free of charge. The new NYSE and NASD rules increase analyst
certification and continuing education requirements. Thus there are plenty of factors that
could have contributed to the decline in analyst coverage that have nothing to do with a
higher Chinese Wall. Supporting the hypothesis that factors unrelated to the Chinese
Wall fostered the decline in research coverage, the Wall Street Journalreports a dramatic
post-regulatory decline in coverage by independent research firms (Eisinger, 2006).
One potential flaw in my analysis of IPO's is that the higher Chinese Wall may
have had an adverse impact on coverage of firms that do not pay IPO underwriting fees.
It is plausible that such fees, during the pre-regulatory period, were necessary to cover the
fixed costs of research. Prohibiting firms from using underwriting fees to purchase
research coverage, therefore, might have reduced coverage of even those stocks that do
not generate fees. Nevertheless, since analyst coverage did not drop to zero in the postregulatory period, the fixed costs of research must still be getting covered in some
sectors. If the Chinese Wall regulations eliminated IPO fees as a channel to purchase
coverage, they would have reduced analysts' marginal benefit of covering a fee-paying
firm more than that of a non-fee payers in sectors where fixed costs are still getting
covered. I can therefore validly infer from my results, which show an equal reduction in
coverage for both fee-payers and non-fee payers, that the new Chinese Wall regulations
do not have this adverse affect.
The passage of the Sarbanes Oxley act around the same time as the raising of the
Chinese Wall introduces another potential problem for my analysis of IPOs. The act
greatly increased the costs of going public (Carney, 2005). Firms that go public in the
post-regulatory period, therefore, are likely different than those in the pre-regulatory
period. Specifically, post-regulatory IPO firms are likely bigger, more established and
more likely to be in industries where accounting is more straightforward. I thus construct
my IPO sample by matching each post-regulatory IPO to a pre-regulatory IPO using
industry, exchange, and market capitalization. It is thus unlikely that changes in IPO firm
characteristics induced by Sarbanes-Oxley are driving my results.
Sarbanes-Oxley, at first glance, appears to present yet another problem for my
analysis of both SEO's and IPO's. The act mandates more stringent audits and increases
the penalties for accounting fraud (Carnie, 2005). Thus it could be argued that demand
for analyst coverage declined in the post-regulatory period because now financial
statements are more trustworthy. This argument, however, is based on a
misunderstanding of the role of a financial analyst. Analysts are generally not successful
at detecting fraud and/or accounting misstatements, as Griffen (2003) documents. Thus it
is unlikely that Sarbanes-Oxley changed the demand for analyst coverage.
The rest of this study is organized as follows. In section 2 1 describe my sample
selection procedure and present descriptive statistics. In section 3, 1 describe my tests
and present results. Section 4 concludes.
2. Sample Selection and Descriptive Statistics
SEO's and IPO's. From the Securities Data Corporation, I gather data on
exchange listing, offer date, and offer price for every IPO undertaken in the US from
1995-2000 and August 2002 - December 2003. The latter is my post-regulatory sample,
and the former is the pre-regulatory. I exclude shareholders rights issues, issues not
underwritten, and issues for which no information on underwriters is available. I exclude
the January 2001- July 2002 period because Spitzer was conducting his investigation
during this time and it is uncertain whether the regulations were already having an effect.
I end the post-regulatory IPO sample in 2003 because I need to match IPO firms to
contemporaneous control firms that do not generate any equity underwriting or M&A
fees for at least two years after the IPO. I also obtain from CRSP data on every SEO
undertaken between 1995 and 2005, excluding the January, 2001-July, 2002 investigation
period, keeping only those SEO's that were public for at least 1 year.
Analyst Coverage. From I/B/E/S I obtain data on analyst forecasts within the 6
month period following an IPO or SEO. I define a dummy variable, coverage, that
indicates whether at least one analyst published a forecast of the firm's earnings during
this period. I also define a variable, #analysts,that equals the number of analysts
publishing at least one forecast during this period. For SEO's, I also obtain #analystsfor
the six to twelve month period prior to the offering date. I then compute the difference
between the post-offering and pre-offering values of #analysts. I use the six to twelve
month pre-SEO period, rather than just the six month pre-SEO period, to compute preSEO analyst coverage because competition for SEO underwriting business may be
affecting analyst coverage during the six month period leading up to the SEO. I
winsorize the change in analyst coverage at the
however, are not sensitive to winsorization.
9 5th
percentiles of -2 and 7. My results,
SEO Control Variables. In my tests, I want to control for variables that affect
analyst coverage. Bhushan (1989) identifies institutional holdings (insthld), the number
of institutional holders (holders), volume, size, return volatility (vol), the firm's squared
correlation with the market (r2), and the number of lines of business (lob) as important
determinants of analyst coverage. Roulstone (2003) identifies measures of liquidity as
being important as well. I am able to obtain measures of each for SEOs. Using CDA
Specturm, I obtain the number of institutions holding shares (holders)and their total
stake in the firm (insthld)as last reported at least six months before the SEO. Both these
values are wonsorized at the
95 th
level to remove outliers. From CRSP, I obtain total
share turnover (turn) during the pre-SEO 6-12 molth period and use it as my proxy for
volume. This variable also serves as a proxy for liquidity. Also using CRSP, I compute
for this period the squared correlation (r2) of the firm's daily returns with the valueweighted aggregate market index. I also obtain for this period the standard deviation of
daily returns (vol), as well as average daily bid-ask spread, normalized by closing price
((ask-bid)/close). The latter serves as another proxy for illiquidity. I use the number of
distinct 3-digit SIC codes of an SEO given in SDC as my proxy for the number of lines of
business (lob), winsorized at the
9 5 th
percentile of 6 to remove outliers. For size, I use
the last available market capitalization, computed from CRSP data, of the 6-12 month
pre-SEO period (mktcap). Finally, following Roulstone (2003), I use a dummy variable
lowprice to indicate whether the price is lower than $15, another proxy for illiquidity. I
use pre-SEO values for all of the above in order to avoid endogeneity bias. All variables
are defined in Table 1.
[Table 1]
IPO Control Variables. IPO's have no data prior to the offering date. Any
control variables, therefore, must be from the post-IPO period. I thus do not use post-IPO
institutional holdings because they are likely affected by post-IPO analyst coverage, and
would therefore introduce endogeneity bias into my regressions. I use the market
capitalization on the day after the IPO as my measure of size. I also use the price from
this date to construct the lowprice dummy. Because analysts cannot cover IPO's during
the 25 day quite period following the offering, variables corresponding to this period are
unlikely affected by analyst coverage. I thus use the average daily bid-ask spread
(quietba)and total share turnover (turn) from this period as controls. In addition, I
compute the percentage change in price from the offer to the next day and use it as a
proxy for underpricing, labeling it up. I also construct a dummy for whether the IPO is
venture backed (ven). Finally, there seems little reason to believe that analyst coverage
should affect daily return volatility (vol) or the squared return correlation with the market
(r2), so I use CRSP to compute vol and r2 during the 6 month post-IPO period and
include them as controls. My results do not change if I omit them.
Table 2 contains descriptive statistics on SEO's in my sample, comparing preregulatory and post-regulatory values. The average increase in the number of analysts
following a firm that completes an SEO declines slightly in the post regulatory period,
but only by 0.2, a negligible amount. Thus the SEO descriptive statistics provide little
support for the hypothesis that the higher Chinese Wall has hurt analyst coverage. The
descriptive statistics also indicate that the post-regulatory SEO firms are also somewhat
larger, less volatile, and more highly correlated with the market.
[Table 2]
Table 3 contains descriptive statistics on SEO's that have no coverage before the
offering. Notice that the proportion that obtain coverage after the offering is actually
higher during the post-regulatory period, contradicting the hypothesis that the Chinese
Wall had an adverse affect on coverage. Not surprisingly, the firms in this sample are
smaller, more volatile, and less correlated than the full sample, described in Table 2.
[Table 3]
Table 4 contains descriptive statistics on all IPOs in the pre and post regulatory
periods. Coverage declined dramatically in the post-regulatory period: only 49% of
IPO's in the post-regulatory period had coverage, as compared to 80% in the preregulatory period. However, the composition of firms changed dramatically as well. The
median post-IPO market cap increased dramatically, from 160 million to 244 million.
The proportion of IPO's listing on NASDAQ declined dramatically, from 84% to 35%.
The number of venture-backed IPO's also declined dramatically, and the industry
composition (not shown) also changed.
[Table 4]
To ensure that differences in pre-and post regulatory IPO firm characteristics do
not drive my results, I match every IPO in the post-regulatory period to the pre-regulatory
IPO that is in the same Fama-French industry, has the same NASDAQ listing status, and
is the closest in post-IPO market capitalization (mktcap). I match without replacement,
so no IPO appears twice in my sample, thereby reducing the potential for serial error
correlation to bias my hypothesis tests. If I am unable to match a pre-regulatory IPO that
has the same Fama-French industry and NASDQ status, I match by NASDAQ status and
market capitalization alone.
Table 5 presents the descriptive statistics for the post-regulatory IPO sample and
the pre-regulatory matched sample. The two samples now appear much more similar.
The median market caps of the two samples are not much closer, differing by less than
$20 million. The proportion of venture and non-venture backed IPO's is the same in both
samples, as is the proportion of firms listing on NASDAQ.
[Table 5]
Next, I construct my control sample. Using CRSP and SDC, for every calendar
month I identify all publicly traded firms that have been listed for at least 2 years and did
not retain an equity underwriter or M&A advisor within two years, before or after the
calendar month. I take this last step in order to ensure that firms in my control sample are
not generating investment banking fees. I then match each IPO to one of these firms
using the same matching procedure as before, using Fama-Fench industry, NASDAQ
status, and market capitalization.
Table 6 presents descriptive statistics for the control sample. Comparing the
control sample and IPO sample, it is apparent that they are similar in all respects but
market capitalization. The average control firm tends to be smaller by just less than $100
million in the post regulatory period, but this difference is less dramatic for the median.
It should not, however, be of great concern since I control for market capitalization in all
my multivariate tests.
[Table 6]
The descriptive statistics indicate that analyst cowrage did in fact decline for new
IPOs. When comparing matched samples of IPO's before and after the regulations, the
difference is less dramatic, but still large: the proportion with coverage dropped from
67% to 49%/o, a difference of just under 20 percentage points. This difference, however, is
similar to that of the control sample, wherein the proportion with coverage dropped from
52% to 35%/0. Thus from the descriptive statistics, it appears that coverage declined for
IPO firms by the same amount as it did for firms that do not generate investment banking
fees.
In summary, preliminary evidence suggests that the higher Chinese Wall did not
have an adverse affect on analyst coverage and other factors. I subject this hypothesis to
more rigorous multivariate tests in the next section.
3. Tests and Results
3.1 SEO's
In this section, I use regression analysis to examine the extent to which the
increase in analyst coverage enjoyed by firms engaging in SEO's declined during the
post-regulatory period. I first estimate the following OLS specification:
chng(#analysts)= 0 lpostreg+P2turn+3baspread+P4mktcap+5vol+P6r2
+P7 insthld+8sholders+910owprice+t9bubble+-e (Model Si)
Where chng(#analysts) is the difference between the number of analysts covering the
firm after the SEO and the number covering it beforehand. If the higher Chinese Wall
had an adverse impact, the coefficient on postreg should be negative. I cluster standard
errors on calendar quarter to ensure they are robust to heteroskedasticity and arbitrary
cross-sectional correlation. Since the data are largely cross-sectional, with firms rarely
appearing more than once, serial correlation is not a concern.
The results of the analysis can be found in Table 7, column 1. Contrary to the
hypothesis that the Chinese Wall hurt coverage, the coefficient on postreg is positive. It
is not, however, statistically or economically significant. Thus I infer that the regulations
did not affect the ability of firms undergoing SEO's to attract analyst coverage. To test
whether these results are robust, I estimate models S2-S4. In Model S2, I add industry
fixed effects, based the Fama French classification system, since Bhushan (1989) finds
that industry is an important determinant of coverage. In Model S3, I restrict the sample
to SEO's that have not had another equity offering or M&A deal for at least three years
so as to isolate the effect of SEO fees on coverage. Finally, in Model S4, I include only
SEO's that had pre-offering market capitalizations smaller than the sample median of
$314 million to test whether the Chinese Wall has an adverse effect concentrated in small
firms. In all of these models, the coefficient on postreg comes in positive, contrary to the
hypothesis that the higher Chinese Wall hurt analyst coverage.
[Table 7]
I also model the extent to which the new regulations hurt the ability of firms
without any research coverage prior to the SEO to obtain it. For this purpose I estimate
the following logistic model on the sample of firms that had no analyst coverage prior to
the SEO:
P(coverage) = P lpostreg+P2turn+P3baspread+P4mktcap+P
5vol+
0 6r2+P7 insthld+fsholders+s8lowprice+Psbubble+E (Model S5)
Where coverage is a dummy indicating that the firm has at least one analyst publish a
forecast within the six months after the SEO. Contrary to the hypothesis, the coefficient
on postreg is positive, and it is not statistically significant. It is also economically small.
A coefficient value of 0.275 implies that the odds of a firm without coverage obtaining it
post-SEO increase by a factor of only 1.31. As a robustness check, I estimate models S6
and S7, which are identical to S5 except that model S6 has industry fixed effects and for
S7 the sample is restricted to firms with market capitalizations below $314 million. The
results are qualitatively similar for all three models. I conclude, therefore, that the new
Chinese Wall regulations have not affected the ability of firms without coverage to obtain
it after an SEO.
3.2 Post-IPO analyst coverage
To tease out the effects of the new Chinese Wall regulations on new IPO's, I use
logistic regression analysis to test whether the log odds of IPO firms receiving coverage
declines with the adoption of the regulations more than it does for the control firms. If
the higher Chinese wall has a negative effect on analyst coverage, it should be stronger
for IPO firms because they pay significant investment banking fees, whereas the firms in
my control sample do not. I thus estimate the following logistic regressions:
odds(coverage)=exp(a+P(postreg+22ipo-•3mktcap+ +P4turn+ 5quietba +P6vol
+P7r2+P8lowprice+P
9 bubble+E) (Model Il)
odds(coverage)=exp(a+plpostreg+P
2postreg*ipo+P3ipo+P
4 mktcap+fPsturn+
(Model 12)
Nsquietba +P7vol +I 8r2 +P9lowprice+P lobubble+E)
Where coverage is a dummy indicating that the firm has at least one analyst forecast
published within 6 months of the IPO. IPO is a dummy indicating that the firm is the
IPO sample and not the control sample. The control variables are defined similarly to
those used in the SEO regressions, but they are computed during the post-IPO period. In
particular, mktcap is the market capitalization on the day after the offering, and quietba
and turn are the average bid-ask spread and total turnover computed during the 25 day
post-IPO quite period. The variables r2 and vol are computed over the 6-month post-IPO
period, and I compute standard errors by clustering on calendar quarter, making them
robust to heteroskedasticity and cross-sectional correlation. Serial correlation is not a
concern since no firm appears more than once in the sample.
Model Il measures direct effects. The parameter on the interaction term in Model
12 measures the extent to which the log odds of coverage declined more for IPO firms
than non-IPO firms in the post regulatory period. If the Chinese Wall regulations have an
adverse effect, the log odds of coverage in the post-regulatory period should have
declined more for IPO firms than for the control firms, implying that the parameter on the
interaction term should be negative. Parameter estimates and their standard errors are
given in Panel A Table 8.
[Table 8]
Consistent with the results in the previous section, the parameter corresponding to
the postreg dummy is negative and significant, meaning that the log odds of coverage
declined for IPO firms and their control firms during post-regulatory period. However,
the parameter associated with the interaction term is indistinguishable from zero,
implying that the change in log odds for both IPO and non-IPO firms was approximately
the same. Thus it appears that the decline in coverage for IPO firms was not the result of
the new Chinese Wall regulations.
To get a sense of the economic significance of the interacted effect, I use the
parameters of Model 12 to estimate the difference in the amount that the probability of
coverage changed in the post regulatory period for IPO firms versus that of non-IPO
control firms. I use the method suggested in Norton and Ai (2003) and Norton, Wang
and Ai (2004). Because of the nonlinearity of the logistic regression, the magnitude of
the interacted effect varies with base probability, i.e. the probability of coverage for a
given firm given ipo=0 and postreg=0. For this reason, I compute the interacted effect
for various base probabilities of coverage: the probability given all control variables are
at their sample mean, the probability given they are at their median, and the unconditional
sample mean of coverage for the pre-regulatory period. Estimates of the interacted effect
and their standard errors are given in Table 9.
[Table 9]
In each case, the difference in change in probability of coverage for IPO firms and
control firms no more negative than -.008, an economically negligible amount. However,
the standard errors of the interacted effects, computed using the delta method, are high,
indicating low power. The low power is driven by the large standard errors of the
parameters associated with quietba, r2, and vol. If I eliminate those variables from the
specification, the estimated interacted effects do not change materially, and all the
standard errors drop to below 0.07. An error of this magnitude implies that if the
interacted effect were at least as negative as -0.12, I would be able to detect it at the 5%
level of significance using a one-sided test. I can thus say with a high degree of
confidence that the probability of coverage of IPO and non-IPO firms declined by the
same amount in the post-regulatory period.
I also analyze the interacted effect on the number of analysts covering a firm. I
run the following OLS specifications:
#analysts-=a+ lpostreg+P2ipo+ 3mktcap+ + 4 turn+ 0squietba +P6vol +47r2
+P8lowprice+p9bubble+4(Model 13)
#analysts=±+p lpostreg+02postreg*ipo+3ipo+f4mktcap+f5turn+pquietba+f7vol
+P8 r2 +p9lowprice+Pjobubble+E(Model14)
where all variables are defined as before. I winsorize #analystsat the
9 9 th
percentile to
remove outliers, but my results are not sensitive to this. As with the logit specifications, I
compute standard errors by clustering on calendar quarter.
The coefficient on the interaction term in Model 14 measures the extent to which
the adoption of the new Chinese regulations has a different effect on coverage for IPO's
than it does for non-IPO's. A negative estimate would imply that coverage declined
more for IPO firms than it did for non-IPO firms, as predicted by the hypothesis that the
Chinese Wall regulations had an adverse effect. Estimates for Models 13 and 14 can be
found in the last two columns of Table 8.
Contrary to the implications of the hypothesis that the higher Chinese Wall
harmed analyst coverage, the estimate of the coefficient on the interaction term in Model
6 is statistically indistinguishable from zero. Its economic significance is also small. A
estimate value of -0.09 implies that coverage of IPO firms declined by a negligible 9
hundredths of an analyst more than that of non-IPO firms in the post-regulatory period.
In addition, the standard error for the interaction term coefficient, 0.26, is also small,
implying the power of my test is high. A standard error of this magnitude implies that I
would be able to detect a difference in coverage change for IPO's versus non-IPO's as
small as 0.50 of an analyst at the 1% level of significance using a one-sided test.
4. Conclusion
This paper tests the hypothesis that the new regulations, adopted in 2002, that
further restrict interaction and cooperation between investment bankers and analysts,
adversely impact analyst coverage of firms that pay investment banking fees. Contrary to
this hypothesis, I find that the increase in analyst coverage that normally accompanies an
SEO did not change with the adoption of new regulations. In addition, while I find that
new IPO firms tend to enjoy less coverage in the post-regulatory period, they enjoy just
as much coverage as a control sample of comparable firms that generate no investment
banking business. If a higher Chinese Wall adversely impacts analyst coverage, it should
affect coverage of firms that generate substantial investment banking business, like new
IPO's, more than that of firms that do not. I therefore conclude from my results that the
higher Chinese Wall does not adversely impact analyst coverage.
Policy makers such as Elliott Spizter have expressed concern over the significant
decline in analyst coverage that came with the adoption of many new regulations over the
2000-2002 period (Parker, 2005). Given the evidence of the beneficial effects of analyst
coverage on liquidity and cost of capital (Brennan and Subrahmanyam, 1995, Roulstone,
2003, Yohn, 1998), devising a means to reverse this decline is an important public policy
question. While this study does not provide an answer, it does provide direction as to
where it may be found. Many other regulations were adopted around the same time as
the new regulations restricting investment barker/analyst cooperation, including Reg FD
as well as more stringent requirements of institutional oversight as well as new analyst
certification and education. By casting doubt to the hypothesis that the higher Chinese
Wall is responsible for a reduction in analyst coverage, my results suggest that that these
other policy changes are the more likely culprits. Thus I provide important guidance for
future research.
References
Barber, Brad Rueven Lehavy and Brett Trueman, 2004. "Comparing the Stock
Recommendation Performance of Investment Banks and Independent Research
Firms." Unpublished working paper. University of California, Los Angeles.
Bhushan, Ravi, 1989.
"Firm Characteristics and Analyst Following."
Journal of
Accounting and Economics 11, 255-274.
Boni, Leslie, 2005. "Analyzing the Analysts after the Global Settlement." Unpublished
working paper, University of New Mexico.
Brennan, Michael and Avanidhar Subrahmanyam, 1995. 'Investment Analysis and Price
Formation in Securities Markets" Journalof FinancialEconomics 38, 361-381.
Carnie, William, 2005. "The Costs of Being Public after Sarbanes Oxley: The Irony of
'Going Private'." Law and Economics Research Paper #05-4, Emory University
School of Law.
Dechow, Patricia M., Amy P. Hutton and Richard G. Sloan, 2000. "The Relation
Between Analysts' Forecasts of Long-Term Earnings Growth and Stock
Performance Following Equity Offerings," Contemporary Accounting Research
17, 1-32.
"The Effect of Investment Banking
Dugar, Amitabh and Siva Nathan, 1995,
Relationships on Financial Analysts' Earnings Forecasts and Investment
Recommendations," ContemporaryAccounting Research 12, 131-160.
Eisinger, Jesse, 2006. "Long & Short: Why Independent Research is Drying Up." Wall
Street Journal(March 8), C1.
Gomes, Armando, Gary Gorton and Leonardo Madureira, 2004 "SEC Regulation Fair
Disclosure, Information and the Cost of Capital." NBER working paper
#W10567.
Griffin, Paul A., 2003.
'A League of Their Own? Financial Analysts' Responses to
Restatements and Corrective Disclosures." Journalof Accounting, Auditing and
Finance 18, 479-518.
Iskoz, Sergey, 2002. "Relative Performance and Institutional Reaction of Underwriter
Analyst Recommendations." Unpublished Working Paper. MIT Sloan School of
Management.
Kadan, Ohad, Leonardo Madureira, Rong Wang and Tzachi Zach, 2006. "Conflicts of
Interest and Stock Recommendations - The Effects of the Global Settlement and
Papers and Proceedings of the American Finance
Related Regulations."
Association Meetings, Boston.
Kolasinski, Adam and S.P. Kothari, 2006. "Investment Banking and Analyst Objectivity:
Evidence from Analysts Affiliated with M&A Advisors." Unpublished working
paper, MIT Sloan School of Management.
Krigman, Laurie, Wayne H. Shaw, and Kent Womack, 2001, "Why Do Firms Switch
Underwriters?" JournalofFinancialEconomics 60, 245-84
Lange, Julian E., 2001. 'Smart Money? The Impact of Having Top Venture Capital
Investors and Underwriters Backing a Venture." Venture Capital3, 309-26
Lin, Hsiou-wie and Maureen F. McNichols, 1998, "Underwriting Relationships,
Analysts' Earnings Forecasts and Investment Recommendations," Journal of
Accounting andEconomics 25, 101-27.
Ljungqvist, Alexander, Felicia Marston and William J. Wilhelm, 2005, "Competing for
Securities Underwriting Mandates: Banking Relationships and Analyst
Recommendations," JournalofFinance, forthcoming.
Malmendier, Ulrike and Devin Shanthikumar, 2004. "Do Securities Analysts Speak in
Two Tounges?" Unpublished working paper, Stanford University.
Michaely, Roni and Kent Womach, 1999, "Conflict of Interest and the Credibility of
Underwriter Analyst Recommendations," Review of FinancialStudies 12, 653-86
Norton, Edward and Chunrog Ai, 2003. "Interaction terms in logit and probit models."
Economics Letters 80, 123-29.
Norton, Edward, Hua Wang, and Chunrog Ai, 2004 "Computing interaction effects and
standard errors in logit and probit models." The Stata Journal4, 103-16.
Parker, Andrew, 2005. "Learning to cope with a three-year blizzard of reform."
FinancialTimes (January 27), 4.
Rock, Kevin, 1986. "Why new issues are underpriced." JournalofFinancialEconomics
15, 187-212.
Roulstone, Darren, 2003. "Analyst Following and Market Liquidity." Contemporary
Accounting Research 3, 551-578.
Yohn, T., 1998. "Information asymmetry around earnings announcements." Review of
QuantitativeFinanceand Accounting 11, 165-182.
Yonca, Ertimur, Jayanthi Sunder and Shyam Sunder, 2005. "Measure for Measure: An
examination of the Association between Forecast Accuracy and Recommendation
Profitability of Sell-side Analysts." Unpublished working paper, Stanford
University.
Table 1
Variable Definitions
Description
Variable
chng(#analysts)
Change in #analysts from the 6-12 month pre-SEO to the 6 month post-SEO period
Dummy indicating that at least 1 analyst issued at least I earnings forecast within 6 months of
the IPO or SEO
#analsyts
Number of analysts who issued at least I earnings forecast within 6 months of the IPO
mkltcap
Market capitalization one day after the IPO at market close, in $millions or as last reported
during the 6-12 month pre-SEO period
nasdq
Dummy inidcating the IPO was listed on NASDAQ
turn
Total share turnover during the 25 day quiet period following the IPO or during the 6-12
month pre SEO-Period
holders
Number of institutional holders as last reported before an SEO
insthld
% of firm owned by institutions as last reported before an SEO
baspread
Average (ask-bid)/close of an SEO firm during the 6-12 month pre-SEO period
quitba
The average (ask-bid)/close during the 25 day quiet period following an IPO
postba
The average (ask-bid)/close during the 6 months an IPO, excluding the 25 day quiet period
lob
Square of the correlation of a firm's returns with the CRSP value-weighted index during the 612 month period before an SEO or the 6-month period after an IPO
Standard deviation of a firm's returns during the 6-12 month period before an SEO or the 6month period after an IPO
Number of 3-digit SIC codes of an SEO firm, as given in SDC
lowprice
Dummy indicating a price < $15
up
% change in the price from the time the offering to the close one day after the offering
ven
Dummy Indicating that the IPO is venture-backed
bubble
Dummy indicating the IPO was undertaken during the bubble period, i.e. 1998-2000.
vol
Table 2
Descriptive Statistics on SEO's Descriptive Statistics on SEO's from 19952000, 8/2002-2005. Variable definitions given in Table 1
chng(#analysts)
mktcap
baspread
turn
holders
insthld
r2
vol
lob
lowprice
Mean
1.898
1439
0.044
0.844
#####
0.354
0.062
0.035
2.425
0.349
Pre-regulatory Sample
Stdev
min
2.290
-2.000
7922
0
0.025
0.008
0.973
0.002
51.302
0.000
0.245
0.000
0.078
0.000
0.019
0.006
1.536
1.000
0.477
0.000
(N=1602)
p25
p50
0.000
2.000
100
244
0.025
0.040
0.295
0.543
13.000 30.000
0.151
0.328
0.008
0.031
0.022
0.033
1.000
2.000
0.000
0.000
chng(#analysts)
mktcap
baspread
turn
holders
insthld
r2
vol
lob
lowprice
Mean
1.688
1957
0.038
0.840
#####
0.461
0.148
0.028
2.558
0.399
Post-regulatory sample (N=913)
Stdev
min
p25
2.240
-2.000
0.000
11200
2
191
0.009
0.021
0.022
0.992
0.011
0.322
63.777
0.000
32.000
0.278
0.000
0.211
0.132
0.000
0.038
0.017
0.007
0.016
1.644
1.000
1.000
0.490
0.000
0.000
p50
1.000
467
0.032
0.567
71.000
0.461
0.114
0.024
2.000
0.000
p75
3.000
729
0.057
1.050
63.000
0.536
0.086
0.044
3.000
1.000
max
7.000
207719
0.271
11.858
219.000
0.867
0.582
0.204
6.000
1.000
p75
3.000
1337
0.048
0.949
125.000
0.712
0.224
0.035
4.000
1.000
max
7.000
314391
0.165
14.557
219.000
0.867
0.711
0.126
6.000
1.000
Table 3
Descriptive Statistics on SEO's from 1995-2000, 8/2002-2005. that had
no pre-SEO analyst coverage. Variable definitions given in Table 1
coverage
mktcap
baspread
turn
holders
insthld
r2
vol
lob
lowprice
Mean
0.761
424
0.057
0.719
8.034
0.101
0.028
0.046
2.315
0.676
Pre-regulatory sample
Stdev
min
0.428
0.000
2398
0
0.034
0.009
1.038
0.010
15.248
0.000
0.152
0.000
0.045
0.000
0.027
0.009
1.503
1.000
0.469
0.000
p50
1.000
58
0.051
0.456
4.000
0.027
0.010
0.042
2.000
1.000
p75
max
1.000
1.000
160 34626
0.072
0.271
0.858 11.435
9.000 ######
0.146
0.867
0.034
0.269
0.057
0.204
3.000
6.000
1.000
1.000
coverage
mktcap
baspread
turn
holders
insthld
r2
vol
lob
lowprice
Mean
0.793
250
0.045
0.711
18.793
0.137
0.057
0.038
2.218
0.690
Post-regulatory sample (N=87)
Stdev
min
p25
p50
0.407
0.000
1.000
1.000
526
2
46
93
0.028
0.009
0.023
0.038
1.205
0.030
0.120 0.288
22.156
0.000
4.000 14.000
0.153
0.000
0.036 0.087
0.082
0.000
0.003
0.025
0.025
0.010
0.020
0.029
1.324
1.000
1.000
2.000
0.465
0.000
0.000
1.000
p75
max
1.000
1.000
202
3778
0.063
0.135
0.681
8.286
22.000 ######
0.185
0.867
0.078
0.487
0.049
0.126
3.000
6.000
1.000
1.000
(N=238)
p2 5
1.000
32
0.033
0.168
1.000
0.000
0.003
0.029
1.000
0.000
Table 4
Descriptive statistics for all IPO's from 1995-2003. Variable definitions in Table 1.
IPO's in pre-Regulatory Period (1995-2000, N=2,905)
mean stdev
min
P25
Median
P75
max
coverage
0.803 0.398
0
1
1
1
1
#analysts 2.738 2.318
0
1
3
4
25
cap
493 1611
2
65
160
390
45121
0.841 0.366
nasdq
0
1
1
1
1
0.357 59.714
0.119
0.007
0.329 1.248 -0.533
up
ven
0.407 0.491
0
0
0
1
1
turn
0.398 0.658
0
0.156
0.27
0.467 25.813
quietba
0.07 0.046 0.003
0.038
0.056
0.089 0.424
postba
0.064 0.035 0.005
0.038
0.056
0.084 0.386
r2
0.054 0.061
0
0.008
0.032
0.077
0.478
vol
0.052 0.026 0.005
0.034
0.047
0.067 0.184
lowprice
0.538 0.499
0
0
1
1
1
IPO's in post-regulatory period (8/2002-2003, N=189)
mean stdev
min
P25
Median P75
max
coverage
0.487 0.501
0
0
0
1
1
#analysts
2.048 2.554
0
0
0
4
14
mktcap
419
593
0
95
244
450
4922
nasdq
0.349 0.478
0
0
0
1
1
0.001 0.063 0.466
0
0.049 0.107 -0.155
up
ven
0.164 0.371
0
0
0
0
1
turn
0.272 0.384 0.012
0.044
0.154
0.37 3.103
quietba
0.023 0.024
0
0.003
0.016 0.038 0.149
postba
0.025 0.019 0.005
0.011
0.017 0.034 0.139
r2
0.033 0.042
0
0.004
0.017 0.045
0.23
vol
0.018 0.014 0.004
0.008
0.013 0.026 0.106
lowprice
0.254 0.436
0
0
0
1
1
Table 5
Descriptive statistics for the matched sample of IPOs. Each post-regulatory IPO (from
August 2002-Dec. 2003) is matched by Fama-French industry, exchange, and market
capitalization (cap) to an IPO in the pre-regulatory period. Variable definitions given in
Table 1.
coverage
#analysts
mktcap
nasdq
up
ven
turn
quietba
postba
r2
vol
lowprice
coverage
#analysts
mktcap
nasdq
up
ven
turn
quietba
postba
r2
vol
lowprice
IPO Matched Sample, Pre-Regulatory Period (N=189)
mean
stdev
min
P25
Median
P75
0.667
0.473
0
1
2.471
2.373
0
0
4
2
357
413
2
105
226
417
0.349
0.478
0
0
0
1
0.119
0.256
-0.2
0
0.128
0.025
0.164
0.371
0
0
0
0
0.286
0.345
0.003
0.077
0.182
0.367
0.038
0.042
0.003
0.012
0.046
0.025
0.039
0.038
0.005
0.018
0.046
0.027
0.06
0.07
0
0.009
0.033
0.087
0.031
0.023
0.005
0.014
0.024
0.039
0.291
0.455
0
0
0
1
IPO's in post-regulatory period (8/2002-2003, N=189)
mean
stdev
min
P25
Median
P75
0.487
0.501
0
0
1
2.048
2.554
0
4
0
419
593
0
244
450
0.349
0.478
0
0
1
0.049
0.107
-0.155
0.001
0.063
0.164
0.371
0
0
0
0
0.272
0.384
0.012
0.044
0.154
0.37
0.023
0.024
0
0.003
0.016
0.038
0.025
0.019
0.005
0.011
0.017
0.034
0.033
0.042
0
0.004
0.017
0.045
0.018
0.014
0.004
0.008
0.013
0.026
0.254
0.436
0
0
1
0
max
1
10
2033
1
1.875
1
3.103
0.306
0.386
0.354
0.114
1
max
1
14
4922
1
0.466
1
3.103
0.149
0.139
0.23
0.106
1
Table 6
Descriptive statistics for the control sample. For each IPO firm, a ontrol firm is
matched based on Fama-French industry, exchange, and size. To qualify for the control
sample, a firm must have been public for at least 2 years and have not engaged in either
an SEO or and M&A deal within 2 years, before or after, of the treatment firm's IPO
date. Variable definitions given in Table 1.
coverage
#analysts
mktcap
nasdq
turn
quietba
postba
r2
vol
lowprice
coverage
#analysts
mkcap
nasdq
turn
quietba
postba
:r2
vol
lowprice
Pre-regulatory Control Sample (N=189)
stdev
min
P25
Median
0.501
0
0
1
3.108
0
0
1
386
2
94
165
0.483
0
0
0
0.285
0.003
0.022
0.056
0.039
0.003
0.012
0.023
0.041
0
0.015
0.026
0.122
0
0.008
0.03
0.024
0.005
0.013
0.022
0.487
0
0
0
Post-regulatory Control Sample (N=189)
mean
stdev
min
P25
Median
0.354
0.48
0
0
0
2.026
4.462
0
0
0
332
427
0
63
206
0.36
0.481
0
0
0
0.204
0.473
0.003
0.031
0.056
0.025
0.029
0
0.004
0.011
0.029
0.027
0.004
0.011
0.016
0.059
0.114
0
0.004
0.018
0.021
0.021
0.002
0.007
0.011
0.36
0.481
0
0
0
mean
0.524
2.058
309
0.365
0.139
0.037
0.039
0.073
0.031
0.381
P75
1
3
331
1
0.145
0.05
0.048
0.078
0.04
1
P75
1
3
388
1
0.163
0.042
0.04
0.061
0.03
1
max
1
22
2033
1
3.103
0.306
0.389
0.899
0.118
1
max
1
34
3344
1
3.103
0.162
0.182
0.985
0.15
1
Table 7
Analysis of the change in coverage that comes with an SEO's from 1995-2000, and
8/2002-2005. Models S1-S3 conduct OLS analysis of the change in the number of
analysts following a firm pre and post SEO. An analyst is coded as following a firm
post-SEO if he publishes at least one forecast in the six month period after the SEO. An
analyst is coded as following the firm pre-SEO if he publishes a forecast between six and
twelve months before the SEO. In Model S1, firms must have been public for at least
one year before the SEO to be included in the sample. Models S2 and S3 have further
sample restrictions noted below. Models S4-S6 conduct logistic analysis of post-SEO
analyst coverage for firms that have no coverage pre-SEO. In Model S4, firms must have
been public for at least one year before the SEO. Further sample restrictions for Models
S6 restricts the sample to small firms.
postreg
turn
baspread
mktcap
stdret
r2
insthld
holders
lowprice
bubble
Change in # Analysts (OLS)
Model S1 Model S2 Model S3+ Model S4'
0.055
0.160
0.344
0.630***
(0.177)
(0.187)
(0.229)
(0.171)
0.406*** 0.330*** 0.205
0.220*
(0.082)
(0.089)
(0.151)
(0.116)
18.465*** 15.508*** 27.504*** 4.972
(8.378)
(6.047)
(5.587)
(5.817)
0.000***
0.000*** 0.000
0.006***
(0.000)
(0.000)
(0.000)
(0.001)
-14.961**
-18.105*** -25.868*** -9.346*
(6.458)
(8.423)
(6.956)
(5.455)
0.452
0.824
3.381***
1.008
(0.936)
(0.947)
(1.618)
(1.278)
-0.227
-0.255
-0.088
0.216
(0.146)
(0.441)
(0.194)
(0.236)
-0.008*** -0.008*** -0.017*** -0.031***
(0.002)
(0.002)
(0.005)
(0.007)
-0.511 *** -0.617*** -0.729*** -0.063
(0.145)
(0.151)
(0.198)
(0.217)
0.019
0.020
-0.224
0.067
(0.193)
(0.201)
(0.313)
(0.150)
Yes
Ind. FE?
No
2515
2509
Observations
0.09
R2 or Pseudo R2 0.05
Robust standard errors in parentheses
Yes
997
0.13
Yes
1246
0.11'
Post-SEO Coverge Dummy (Logit)
For Firms with no pre-SEO coverage
Model S5 Model S6 Model S7+
0.275
0.386
0.102
(0.349)
(0.387)
(0.350)
0.160
0.102
0.138
(0.225)
(0.196)
(0.204)
7.419
7.251
6.592
(12.961)
(13.154) (12.286)
-0.000
-0.001*
0.009*
(0.000)
(0.000) (0.005)
-21.749
-24.529
-18.506
(16.646) (16.936) (15.795)
-0.443
0.192
-1.117
(3.318)
(3.555)
(4.098)
1.513
0.741
2.247
(1.026)
(1.114)
(1.503)
-0.010
-0.003
-0.019
(0.012)
(0.037)
(0.012)
-0.150
-0.562
0.186
(0.310)
(0.324)
(0.408)
0.492
0.455
0.209
(0.337)
(0.378)
(0.371)
No
325
0.04
Yes
274
0.11
* significant at 10%; ** significant at 5%; *** significant at 1%
+SEOs of firms without IPO, another SEO, or M&A activity within previous 3 years
"Mktcap < $314 million (sample median)
No
280
0.07
Table 8
Regression analysis modeling the research coverage of IPO firms and
control firms. Logit analysis is used for specifications with the coverage
dummy as the dependent variable. Ordinary Least Squares is used for the
specifications with the number of analysts as the dependent variable.
Sample of post-regulatory IPO's (8/2002-12/2003) matched to preregulatory IPO's (1/1995-7/2002) based Fama-French industry, exchange,
and market capitalization (cap). Each IPO, in turn, is matched on the same
characteristics to a control firm which did not pay any equity underwriting
or M&A fees within a four year window around the IPO date.
Independent variable definitions given in Table 1. Dependent variables
include a coverage dummy, indicating at least 1 analyst covered the stock
within 6 months of the IPO, as well as the total number of analysts that
covered the stock.
Predicted
Sign
Test Variables
postreg
-
ipo
+
Coverage Dummy (Logit)
Model 12
Model I1
-1.022***
(0.360)
0.600***
(0.136)
-0.464
(0.281)
0.096
(0.127)
-0.416
(0.316)
0.144
(0.214)
-0.096
(0.254)
0.003***
(0.000)
(0.000)
turn
1.065**
1.404***
(0.366)
(0.522)
(0.522)
41.724
vol
41.657
24.091**
(35.174)
(35.150)
(11.415)
r2
-0.779
-0.777
0.340
(1.194)
(1.117)
(1.113)
lowprice
0.162
0.163
0.200
(0.292)
(0.279)
(0.294)
5.911
5.867
0.634
quietba
(22.741)
(22.792)
(6.100)
-1.001***
-0.999***
bubble
-0.586*
(0.327)
(0.325)
(0.317)
756
756
756
Observations
R2 or Pseudo-R2
0.24
0.24
0.29
Robust standard errors in parentheses, clustered by calendar quarter
0.003***
postreg*ipo
Control Variables
mktcap
0.002***
-1.054**
(0.412)
0.566**
(0.244)
0.066
(0.276)
# Analysts (OLS)
Model 13
Model 14
0.002***
(0.000)
1.068**
(0.000)
1.401***
(0.367)
24.038**
(11.485)
0.332
(1.191)
0.200
(0.279)
0.650
(6.121)
-0.586*
(0.317)
756
0.29
Table 9
Marginal effect of regulation on probability of coverage for IPO firms less the
marginal effect for control firms. Computed using logit parameter estimates,
including interaction terms, given in Table 6. Standard errors computed with
the delta method. Difference in marginal effects computed for various values
of the control variables: their sample mean, sample median, or set such that
the predicted base probability of coverage (where ipo=0O and postreg=O)
equals the mean of coverage during the pre-regulatory period.
Regulatory
Effect For
Control variables set to mean
Control variables set to median
Control Variables set such that
base probability = pre-regultory mean
Base
IPO's - Effect
Standard
Probability
0.60
for non-IPOs
0.027
Error
1.516
0.47
-0.008
1.473
0.59
0.0259
1.522
Chapter 3 (Joint with SP Kothari)
Investment Banking and Analyst Objectivity:
Evidence from Analysts Affiliated with M&A Advisors
1. Introduction
In the wake of the market downturn in 2000, regulators, financial commentators,
plaintiffs in shareholder lawsuits, and others alleged that analysts working for investment barks
have a conflict of interest and bias their research at the behest of bankers and banking clients.
Over the course of 2001 and 2002, New York State Attorney General Elliot Spitzer uncovered
internal bank documents suggesting that a number of analysts compromised their integrity in this
manner. The ten largest investment banks settled with Spitzer in 2003 (the "Spitzer Global
Settlement").
They agreed to prohibit analysts from working on investment banking deals,
remove the generation of investment-banking revenue as a factor in analyst compensation, and
disclose in analyst reports any investment banking relations, including M&A relations. 9 In July
of 2002, both the NASD and NYSE adopted new rules incorporating many elements of the
Spitzer Global Settlement.20
In an attempt to verify whether the isolated cases uncovered by regulators and
prosecutors are indicative of a widespread phenomenon, numerous studies test whether conflicts
arising from equity underwriting relations lead analysts compromise their objectivity. 2 1 In
contrast, we study conflicts arising from M&A relations, which have also concerned regulators,
but up to now have largely been ignored in the literature.
Using a sample of analyst
'9 Chronology of the Merrill Lynch Probe." The Associated Press. May 21, 2002. Also see the reports of the New
York State Attorney General on his website: http://www.oag.state.ny.us/.
20 See NASD rule 2711 and NYSE rule 472. See also NASD and NYSE (2005) for a summary of the new rules, as
well as a history of the events leading up to their adoption, as well as a comparison of their provisions with that of
Spitzer's Global Settlement with the 10 Investment Banks.
See Dechow, Hutton and Sloan (2000), Dugar and Nathan (1995), Iskoz (2002),
Lin and McNichols (1998),
Malmendier and Shanthikumar (2004), Michaely and Womack (1999) and O'Brien, McNichols, and Lin (2005).
21
recommendations issued around M&A deals from 1993-2001, we find evidence that M&Arelated conflicts significantly influence analyst recommendations.
Analysts affiliated with
acquirers are more likely than unaffiliated analysts to upgrade their recommendation of the
acquirer around M&A deals, including all-cash deals. We also find that target-affiliated analysts
tend to upgrade acquirers in all-stock deals after exchange ratios are set, thereby benefiting target
managers and shareholders. For reasons discussed below, it is unlikely that selection bias can
explain these findings, leaving us to conclude they are the result of conflict of interest.
We believe these results are important for at least three reasons. First, analyst conflicts
arising from M&A relations are likely a more pervasive phenomenon than those arising from
equity underwriting relations. The number of firms engaging in M&A transactions is vastly
greater than the number issuing equity in each year we study. Also, M&A fees exceed equity
underwriting fees in each of these years (see section 2). Second, practitioners and regulators
have expressed concern that M&A relations were creating analyst conflicts. The Spitzer Global
Settlement requires analysts to disclose whether the subject of a report is their employer's M&A
client (Goff, 2002). Reingold and Reingold (2006, Chapters 5 & 6) document instances of
analyst behavior around M&A deals that led some practitioners to suspect that analysts were
tainting their research to benefit M&A clients. 22 Our analysis confirms that this problem was not
limited to a few anecdotes. Finally, in the M&A context we are able to empirically identify the
effect of conflict of interest and distinguish it from selection bias, a goal that previous research
on analyst conflicts has not been entirely successful in attaining.
22 Deals around which Reingold and Reingold allege affiliated analysts compromised their research to benefit M&A
clients include WorldcomMCI, SBC-Ameritech, and Frontier-Global Crossing.
Numerous previous studies find that analysts affiliated with equity underwriters publish
reports about client firms that are more optimistic than those published by unaffiliated analysts. 23
These findings, however, do not necessarily indicate that there is a conflict of interest problem,
as selection bias can also explain these results. It is plausible that having an analyst optimistic
about the issuer enhances an underwriter's ability to execute an equity offering.
Hence,
underwriters with honestly optimistic analysts may be more likely to get selected to do an
underwriting deal, resulting in an innocuous association between analyst optimism and affiliation
with the equity underwriter.
In the M&A context, however, there are circumstances wherein
selection bias is unlikely. As a result, the interpretation of our evidence is less ambiguous.
Specifically, selection bias is unlikely to explain our finding that analysts affiliated with
the acquirer advisor in all-cash deals tend to upgrade the acquirer. The acquirer's stock price is
irrelevant in a cash deal, so the acquirer-affiliated analyst's opinion is irrelevant to the analyst's
employer's execution ability. Hence, cash acquirers are unlikely to select advisors on the basis
of analyst optimism except to reward such optimism. Thus, conflict of interest is the likely
explanation of out result. In addition, selection bias is unlikely to explain our finding that targetaffiliated analysts tend to upgrade acquirer stock after, but not before, the exchange ratio of a
stock deal is set. It is difficult to envision how honest pre-deal analyst priors, which targets
could use to select advisors, would result in such conveniently timed upgrades. Such upgrades,
however, benefit the target, making them likely to be the result of conflict of interest.
We also study near-term EPS and long-term growth forecasts. Consistent with prior
research in the equity underwriting context (Dechow, Hutton, and Sloan, 2000 and Malmendier
and Shanthikumar, 2004), we fail to find evidence that analyst affiliation affects near-term EPS
See Dechow, Hutton and Sloan (2000), Dugar and Nathan (1995), Iskoz (2002), Lin
and McNichols (1998),
Malmendier and Shanthikumar (2004), Michaely and Womack (1999)
23
forecasts. We find a statistically significant effect for long-term growth forecasts, but as Lin and
McNichols (1998) find in the equity underwriting context, the economic magnitude is small.
Prior research. Our results complement Ljungqvist, Marston, and Wilhelm (2005).
They find that publication of optimistically biased analyst reports about an issuer or its parent
does not increase the probability of an investment bank winning an underwriting mandate.
These results suggest that selection bias may not fully explain the previously-documented
optimism of underwriter-affiliated analysts, indicating that conflict of interest plays a role. We
bolster this inference by documenting that analysts behave in a manner consistent with conflict
of interest even in situations were selection bias is unlikely.
Three other studies, Bradshaw, Richardson, and Sloan (2003), O'Brien, McNichols, and
Lin (2005), and Malmendier and Shanthikumar (2004), also conclude that analyst conflict of
interest plays a role in equity underwriting.
We believe, however, that plausible alternative
hypotheses can explain their results. We now discuss each in turn.
Bradshaw, Richardson, and Sloan (2003) find that analyst consensus forecasts and
recommendations improve around the time of equity and debt issuances. They argue that a large
portion of analysts bias their forecasts during such times in a bid to win underwriting business,
driving up the consensus. A plausible alternative hypothesis is that, as Baker and Wurgler
(2002) suggest, firms might be strategically timing securities issuances to coincide with periods
of positive market sentiment reflected in analysts' favorable forecasts.
O'Brien, McNichols, and Lin (2005) find that underwriter-affiliated analysts are slower
to downgrade client firms. However, analysts with more honestly bullish priors about a firm
would be slower to downgrade it, and such priors might influence underwriter selection. Hence,
selection bias as an explanation of this result cannot be ruled out
Malmendier and Shanthikumar (2004) find that underwriter-affiliated analysts are
optimistic relative to consensus in their recommendations but not near-term EPS forecasts. This
finding is indeed consistent with conflict of interest. Analysts are likely more willing to taint
their recommendations than near term EPS forecasts. Recommendations are more difficult to
objectively evaluate, and EPS forecasts tend to be geared more toward sophisticated institutional
investors likely to see through any conflict of interest (Malmendier and Shanthikumar, 2005, and
Mikhail, Walther and Willis, 2005).
The above finding, however, is also consistent with
selection bias. It is plausible that analysts' honest opinions abott long-term prospects tend to
vary more than their opinions about the short-term. Thus under selection bias, one expects a
greater difference of opinion between affiliated and unaffiliated analysts about long-term
prospects than the short term. If analysts make their recommendations based on their long-term
as well as short term expectation of financial performance, selection bias would imply that
affiliated analysts would be more optimistic relative to unaffiliated analysts in their
recommendations than in their near-term EPS forecasts.
Some research fails to find evidence of affiliation biasing analyst research. Cowen,
Groysberg, and Healy (2003) as well as Agrawal and Chen (2004) find that analysts employed
by investment banks issue forecasts statistically identical to those issued by analysts at pure-play
brokerage or independent research firms. Barber, Lehavy and Trueman (2004), however, find
that the "buy" and "strong buy" ratings of investment-bank-employed analysts tend to
underperform those of other analysts. It is difficult to draw inferences about analyst conflicts
from these results, however, because most firms covered by investment bank-employed analysts
are not clients or potential clients.
A potential concern about our research design is that all analysts might bias their reports
in a (generally unsuccessful) attempt to win investment banking business (see Bradshaw,
Richardson, and Sloan, 2003). Thus, a comparison of the reports of analysts whose employers
actually win M&A business to analysts who do not might be biased against finding evidence of
conflict of interest. For this reason, we limit our analysis to reports issued within 90 days of an
M&A deal, a time in which M&A advisors are likely to have already been retained.
The
relatively small 90 day window also minimizes selection bias and thus facilitates a relatively
clean test of the conflict of interest hypothesis.
Outline ofPaper. Section 2 discusses the incentives of affiliated analysts and investment
banking clients in the M&A context and describes how we can cleanly identify the effects of
conflict of interest in this setting. Section 3 describes our data and presents descriptive statistics
and preliminary analysis based on such statistics. Section 4 describes our ordered logit analysis
of recommendation upgrades and downgrades and presents results. Section 5 describes our
ordinary least squares analysis of growth forecasts and presents the results. Section 6 concludes.
2. M&A Context to Study Analyst Objectivity
In this section we discuss two topics. First, in section 2.1, we discuss the ways in which
conflicts of interest related to M&A deals might tempt analysts to taint their research. Second, in
section 2.2, we discuss selection bias in the M&A setting and show that there are at least two
instances in which selection bias is highly unlikely. We can thus empirically identify the effect
of conflict of interest.
2.1 Conflict of interest in the M&A context
M&A fees are an important source of revenue for investment banks. According to
Freeman & Co. estimates, and as illustrated in figure 1, in every year since 1994, M&A fees in
the US have been at least as large as equity underwriting fees, and in recent years significantly
larger. In addition, whereas equity issuances are relatively rare events in the life of the firm,
most firms engage in multiple M&A transactions, giving them the opportunity to offer repeat
business in exchange for desired analyst coverage. The desire for more M&A fees, therefore,
constitutes a potentially significant source of analyst conflicts.
[Figure 11
During our sample period, the regulatory environment left plenty of opportunities for
analysts to bias their research to benefit M&A bankers and clients. After October 24, 1997,
affiliated analysts were free to publish reports on M&A clients and counterparties at any time, so
long as they did not disclose any confidential information about the deal. 24 Even before this
date, however, there were still opportunities for analysts to publish research. There were no
restrictions on analysts publishing research after the consummation of a deal. In addition, an
affiliated analyst could publish reports on M&A clients or counterparties before the deal
announcement so long as the report contained no confidential information about the deal
(Reingold and Reingold, p. 33).
The pre-1997 rules, however, were murky about analysts
publishing between announcement and consummation dates.
Some banks' compliance
departments interpreted them conservatively and restricted their analysts from publishing during
this period (Reingold and Reingold, pp. 38-39). Other banks, however, took a more aggressive
view. Consistent with the aggressive interpretation of the rules, we find that, in the period before
October
1997, M&A advisor-affiliated
analysts published 498 reports between
deal
announcement and consummation.
24 On Oct. 24, 1997, the SEC issued a no-action letter declaring it would not enforce regulations in such a manner as
to prohibit M&A-affiliated analysts from publishing reports while deals were pending. The letter and inquiry to
which it responds are available from the SEC as, "NO-ACT, NAFT WSB File No. 111797017, Merrill Lynch,
Pierce, Fenner & Smith Inc., (Nov. 20, 1997)." See also Reingold and Reingold, pp. 163-164.
Panel A of Table 1 summarizes analysts' incentives to taint their
research
to
benefit
M&A clients or bankers. Under the conflict of interest hypothesis, analysts affiliated with the
acquirer advisor publish optimistically biased reports in order to benefit bankers and acquirer
clients. Optimistic research is likely to benefit the parties in question in at least three ways.
First, in a stock-for-stock transaction, acquirers use their stock as acquisition currency. An
acquirer-affiliated analyst's optimistically biased coverage of the acquirer can potentially
increase the value of the acquisition currency and thereby sweeten the terms of deal for the
client. A better deal for the client, in turn, may result in higher fees for bankers. Second, analyst
optimism about the acquirer around the time of the deal might increase the chances of obtaining
shareholder approval for the deal, which benefits both the bankers and managers.
Finally,
optimistic coverage is associated with stock price appreciation (Womack, 1996), which
favorably impacts management compensation. Grateful managers, in turn may reward bankers
with more deals in the future.
[Table 1, Panel Aj
The incentive for acquirer-affiliated analysts to taint their research about the acquirer are
likely stronger in a stock deal, since in this case the analyst might affect the deal terms.
Nevertheless, incentives for optimistic bias remain even in cash deals because such bias may
increase the chances of deal approval as well as satiate an acquirer manager's general desire for
positive coverage.
There are some exceptions to managerial desire for optimistic coverage, however. Since
the market tends to reward managers for beating analyst EPS forecasts, managers do not want
such forecasts to be excessively high (see Bartov, Givoly and Hayn, 2002 and Kasznik and
McNichols, 2002). This consideration, however, does not apply to long-term growth forecasts
and recommendations, neither of which provides a clear benchmark by which to measure
performance. Managers may also want pessimistic coverage just before a stock option grant in
the hope that such coverage will result in a lower strike price. Such events, however, occur with
limited frequency. Managers, therefore, have an incentive to procure optimistic analyst coverage
most of the time, though the desire for analyst optimism is dampened in the case of near-term
EPS forecasts.
Pessimistic coverage about the target might lower the target's stock price, which can
provide an incentive for acquirer-affiliated analysts to publish pessimistically biased research
about the target. On the other hand, pessimistic coverage of the target might make the deal
appear less favorable to acquirer shareholders, which, in turn, may increase the probability that
the deal will not get completed and the acquirer advisor not get its fees. Thus, the predictions of
the conflict of interest hypothesis in this context are unclear.
If optimistic amlyst coverage raises the share price, then optimistic coverage of a target
firm could potentially increase the acquisition price, which benefits both target managers and
M&A advisors.
The conflict of interest hypothesis, therefore, predicts that target-affiliated
analysts will be optimistic about the target.
In stock deals, target mangers and M&A advisors want to ensure that as many shares of
acquirer shares as possible are exchanged for each target share. Thus before the exchange ratio
has been set, target-affiliated analysts have an incentive to publish pessimistic reports about the
acquirer. Such reports could potentially lower the acquirer stock price, thereby making it easier
for the target advisor to demand more acquirer shares. After the exchange ratio has been set,
however, target managers and shareholders are poised to become acquirer shareholders. Thus
target-affiliated analysts, if they are seeking to benefit target managers, have an incentive under
the conflict of interest hypothesis to optimistically bias their reports about the acquirer after
exchange ratio is set. Optimistic research about the acquirer may also increase the odds that
shareholders approve the deal, which in turn increases the odds bankers earn their fees.
2.2. Selection Bias andIdentifying Conflict of Interest
While the preceding discussion suggests a potential for conflict of interest, we
acknowledge that selection bias is conceivable in some circumstances in the M&A context. The
acquirer advisor's goal is to obtain as bw a valuation of the target as possible, so it is plausible
that its execution ability would be compromised if it employs an analyst who is bullish on the
target. Hence, investment banking firms employing analysts with bullish priors about the target
might be less likely to get selected as advisors by the acquirer. Thus, selection bias predicts that
analysts affiliated with acquirer advisors will be pessimistic about the target. For similar
reasons, the target advisor's execution ability might be compromised if it employs an analyst
bearish on the target. Hence the selection bias hypothesis predicts that target-affiliated analysts
will be optimistic about the target.
In a stock deal, the acquirer advisor seeks a high valuation of the acquirer stock, which
may be difficult if the advisor employs an analyst bearish on the acquirer. Thus, selection bias
predicts that acquirer-affiliated analysts will be optimistic about acquirers in stock deals. In
contrast, in a stock deal, the target advisor attempts to argue for a low valuation for the acquirer,
since it wants target shareholders to receive as many acquirer shares as possible. Hence, the
target advisor's execution ability might be compromised if it employs an analyst who issues
optimistic reports about the acquirer before the deal. Thus, selection bias predicts that, in stock
deals, target-affiliated analysts will be pessimistic about the acquirer. There are, however, two
circumstances wherein selection bias is highly unlikely, which we now discuss.
Analysts affiliated with the acquirer advisor in cash deals reporting on the acquirer.
Table 1, panel B, row 1 summarizes the predictions of the conflict of interest and selection bias
hypotheses in this circumstance.
The acquirer's stock price is irrelevant in a cash deal.
Therefore, optimism or pessimism on the part of analysts in the acquirer advisor's employ is
unlikely to affect the latter's execution ability. Thus, it is unlikely that cash acquirers would
select advisors on the basis of analyst optimism for any reason other than to reward biased
research. Hence selection bias would be an unlikely explanation for acquirer-affiliated analyst
optimism about the acquirer around all-cash deals.
On the other hand, acquirer managers'
general desire for optimism, as well as the possibility that positive research may help them sell
the deal to their shareholders, creates incentives for acquirer-affiliated analysts to optimistically
bias their reports about the acquirer. Therefore, finding that acquirer-affiliated analysts in cash
deals are optimistic about the acquirer would constitute evidence in favor of conflict of interest. 25
[Table 1, Panel B]
Analysts affiliated with the target advisor in stock deals. Table 1, Panel B, rows 2 and 3
illustrates the predictions of the selection bias and conflict of interest hypotheses in this
circumstance. In a stock deal, the target seeks to gain as many acquirer shares as possible in
exchange for its own. The target advisor, therefore, may sweeten the deal for the target by
arguing for a low valuation of the acquirer. If having an analyst bearish on the acquirer in its
employ aids the advisor in this task, therefore, conflict of interest and selection bias predict that
25 Sometimes acquirer advisors provide additional services that might introduce selection bias. Employing
an
analyst bearish on the acquirer might compromise the advisor's ability or willingness to arrange or provide financing
for the deal, so selection bias might explain the optimism of analysts affiliated with acquirer advisors who also
arrange or provide financing. Likewise, employing an analyst bearish on the acquirer might compromise the
advisor's ability to deliver a convincing fairness opinion. To ensure that such selection bias does not taint our
analysis, we exclude analysts affiliated with acquirer advisors who either arrange or provide financing or deliver a
fairness opinion.
analysts employed by target advisors in stock deals should be pessimistic about the acquirer
before the deal.
After the share exchange ratio is set, however, target managers are poised to become
acquirer shareholders. If target-affiliated analysts seek to ingratiate target managers, they have
an incentive to be optimistic about the acquirer after, but not before, the exchange ratio is set.
Conflict of interest, therefore, predicts that analysts affiliated with target advisors on stock deals
will be optimistic about the acquirer after, but not before, the ratio is set. It seems unlikely that
selection bias could explain such a finding. If targets were selecting advisors based on analysts'
honest priors about the acquirer, we would not expect optimism in target-affiliated analysts'
recommendations and forecasts about the acquirer to depend on whether the report is published
before or after the setting of the exchange ratio. Thus, finding that target-affiliated analysts
become more optimistic about the acquirer after, but not before, the setting of the ratio would
constitute credible evidence in favor of the conflict of interest hypothesis.
One might argue that once the ratio is set, target-affiliated analysts have little incentive to
taint their reports to benefit target managers since fees may have already been negotiated.
Receipt of fees, however, is typically contingent on deal completion, and optimistic reports on
the acquirer might increase the probability of deal approval by both target and acquirer
shareholders. In addition, M&A advisors may want business from other targets in the future. If
they establish a reputation for generating positive analyst coverage of acquirers after stock
transactions, future target managers will be more likely to hire them.
3. Data and Descriptive Statistics
M&A Deals. We obtain M&A transaction data from Securities Data Corporation (SDC)
for years 1993 to 2001. Our sample solely consists of statutory mergers and acquisitions of
assets. 26 Thus we exclude from our sample buybacks, acquisitions of certain assets, acquisitions
of partial interest, recapitalizations, spin-offs, split-offs, exchange offers, and acquisitions of
remaining interest because analyst incentives in such deals are unclear. We also limit the sample
to deals in which either the target or acquirer or both are public and at least one advisor has been
retained by the target or the acquirer. After applying these criteria, we are left with 9,840 deals
in our sample.
Merging Deals with recommendation upgrades and downgrades. We focus on changes
in recommendations, rather than recommendation levels, because previous research finds that
changes are more economically meaningful (Womack, 1996).
From I/B/E/S we obtain all
available analyst recommendations about targets and acquirers issued within 90 days of the
M&A announcement date. Then, for each analyst, we obtain the last available recommendation
issued before the 90 day period but within two years of the announcement date. We define a
categorical variable, upgrade, and set it equal to 1 if the analyst's recommendation improved, 0 if
it did not change, and -1 if it changed for the worse (i.e. a downgrade).
If, for a particular
analyst, upgrade cannot be computed because of a lack of a prior recommendation, we discard
the observation from the sample. The above criteria leave 6,822 deals in the upgrade/downgrade
sample. We also analyze recommendation levels separately, without the results being tabulated
in the paper, and find the results are qualitatively similar to our analysis of changes.
Merging Deals with Analyst Forecasts. From I/B/E/S we obtain all available long-term
growth forecasts for all acquirers, targets, and their immediate and ultimate parents (as defined
by SDC) published within 90 days of each of the 9,840 deals in our sample.
Of these
26 To implement our sample selection, include only SDC deals in which the field "form of deal" is labeled
as 'AA'
or 'M.' These labels correspond to acquisition of assets and statutory mergers, respectively. This method of sample
selection is the same as excluding deals whose "form of deal" field is labeled 'A', 'AC', 'AR', 'AP', 'R', 'B', and
'EO,' which correspond to spin-offs, acquisitions of remaining interest, acquisitions of certain assets, acquisitions of
partial interest, recapitalizations, buybacks, and exchange offers.
transactions, only 3,330 had at least two long-term growth forecasts of either the target of
acquirer. We require at least two forecasts because we need to compute each analyst's deviation
from consensus. We focus on long-term growth forecasts because previous research has failed to
find an association between investment banking affiliation and analyst optimism in near-term
EPS forecasts (Dechow, Hutton and Sloan, 2000, Malmendier and Shanthikumar, 2004).
Nevertheless, we analyze EPS forecasts separately, without the results being tabulated in the
paper, and fail to find evidence consistent with either conflict of interest or selection bias, i.e., we
fail to find evidence of EPS forecast bias.
M&A Deal Characteristics. Table 2 presents descriptive statistics on the value of the
M&A deals in the long-term growth forecast and recommendation samples. The average deal in
recommendation sample tends to be smaller, but this is not surprising since many firms in the
I/B/E/S database have recommendations but not growth forecasts, and firms with growth
forecasts tend to be larger. Consistent with the stylized facts about M&A, stock deals tend to be
larger than cash deals, and in the broader recommendation sample, all-cash deals are fewer in
number than deals in which some stock is used (the sum of all-stock deals and mixed deals).
[Table 21
Descriptive Statistics on Analyst forecasts and recommendations. Table 3, Panel A
contains the descriptive statistics on upgrades and downgrades in our final sample. Somewhat
surprisingly, analysts tend to downgrade stocks around the time of an M&A deal more often than
they tend to upgrade them. Panel B presents descriptive statistics on growth forecasts. The
mean long-term growth forecast is roughly 20%, qualitatively similar to those found in other
studies.
[Table 31
Analyst Affiliation. Next we determine the affiliation of each analyst who issued a
forecast or recommendation in our sample. This task is not complicated in principle. SDC lists
all M&A advisors retained on a deal, and I/B/E/S provides the name of the securities firm, which
it calls the "broker," employing each analyst issuing a forecast or recommendation.
Unfortunately, the SDC codes for M&A advisors and I/B/E/S codes for brokers are different, and
there is no mapping between the two coding systems. Hence, we individually match I/B/E/S
brokers and SDC advisors by hand using their corporate names. 2 7 Finally, to eliminate potential
sources of selection bias, we delete from our sample any analysts affiliated with acquirer
advisors that provide or arrange financing for the acquirer or deliver a fairness opinion.
Measuring optimism or pessimism in affiliated analyst growth forecasts. In our tests of
forecasts, we seek to determine whether growth forecasts issued by analysts affiliated with M&A
advisors ("affiliated analysts") are optimistic or pessimistic relative to consensus. We take the
following two precautions in calculating the consensus. First, we do not want any forecast in our
consensus to be contaminated by M&A affiliation. Hence we exclude from consensus any
forecasts issued by analysts affiliated with an M&A advisor that was retained within one year of
the forecast date by the firm whose growth is being forecasted, the counterparty to the M&A
transaction, and the parents or subsidiaries of such firms. Second, recent research indicates that
herding behavior may be economically significant. Scharfstein and Stein (1990) initiated the
herding literature with their model of firm manager herding, which Trueman (1994) applies to
analysts. Hong, Kubik and Solomon (2000) as well as Welch (2000) find evidence that analysts
do indeed exhibit herding behavior. Hence to make sure that herding by unaffiliated analysts
27 In most instances, the names in the two databases are qualitatively the same and can be matched
by sight. To
detect parent-subsidiary or common parent relationships, we consult Hoovers Online, the Directory of Corporate
Affiliations, Lexis -Nexis, and corporate websites. Our method only allows us to detect such relationships that
continue into the present, but given that our sample begins in 1993, this problem is unlikely to be serious.
does not taint our estimate of the unaffiliated consensus, we exclude from our consensus estimate
any forecast issued after the one issued by the affiliated analyst.
To calculate the consensus, we average all the unaffiliated forecasts for a given firm
issued within a calendar month and before the affiliated analyst's.
We then calculate the
difference between the affiliated analyst's forecast and the consensus. We also calculate each
unaffiliated analyst's deviation from consensus in the same manner, except we define consensus
in this case as the average of all other unaffiliated analysts' forecasts.
Our results are not
sensitive to alternative definitions of the consensus, e.g., median of all other unaffiliated
analysts' forecasts.
Date exchange ratio fixed. Our tests of the conflict of interest hypothesis require
comparing target-affiliated analyst reports published after the date on which the exchange ratio
of a stock deal is fixed to reports published before this date. Unfortunately, the SDC database
does not make this date available. However, in their analysis of mergers during the 1994-2000
period, Mitchell, Pulvino and Stafford (2002) find that 78% of the time the exchange rate is fixed
on the M&A announcement date. We therefore use the announcement date as a proxy for the
exchange ratio date.
Descriptive statistics on deviationfrom consensus forecasts and upgrades. Panel A of
Table 4 reports descriptive statistics for upgrades by affiliated analysts in various subsamples.
Recall that upgrade takes a value of one if the analyst changes his recommendation for the better
around the time of the deal, takes the value of -1 if the analyst changes it for the worse, and takes
the value of zero if the recommendation remains unchanged.
Panel B presents affiliated
analysts' deviation from consensus long-term growth forecasts. The statistics are presented for
subsamples sorted by the currency used in the deal, analyst affiliation, the target or acquirer
status of the firm upon which the analyst is reporting, as well as whether the report was issued
before or after the M&A transaction. Panel B of Table 4
[Table 4]
The results in Panel A provide some support for the conflict of interest hypothesis. The
mean value of upgrade for acquirer-affiliated analysts covering acquirers around cash deals is
positive and statistically significant at the 1% level. In addition, the statistics indicate that
acquirer-affiliated analysts are more likely to downgrade targets, consistent with both the
hypotheses in the model. However, it does not appear as though target affiliation has much
effect on analyst objectivity, as the mean value of upgrade is statistically indistinguishable from
zero for all target-affiliated subsamples.
The results in Panel B provide weaker support for the conflict of interest hypothesis. In
cash deals, acquirer-affiliated analysts tend to issue growth forecasts about the acquirer that are
close to a percentage point more optimistic than the consensus. The economic significance of
this point estimate is modest, and it is only significant at the 10% level under a one-sided test.
The rest of the statistics in the table, however, fail to detect any significant effect of M&A
affiliation on long-term growth forecasts. This lack of significance, however, is likely due to low
power, since the number of observations is very low.
Thus far, with our univariate tests, we find limited evidence of an association between
acquirer affiliation and optimism about the acquirer, especially in cash deals, thus providing
some support for the conflict of interest hypothesis.
In the next section, we subject our
hypotheses to multivariate tests, which are more powerful and take into account factors such as
deal size that might enhance or dampen the effect of conflict of interest.
4. Ordered Logistic Regression Analysis of Upgrades and Downgrades
In this section we use logistic regression analysis to determine whether there is an
association between M&A advisory affiliation and analyst recommendation changes around
M&A deals. We use logistic regression analysis, rather than least squares, because
recommendation changes are discrete and ordinal. In sections 4.1 we examine the effects of
acquirer affiliation. In Section 4.2 we examine the effects of target affiliation. In all cases, we
compute standard errors by clustering on calendar year to ensure robustness to heteroskedasticity
and arbitrary cross-sectional and intra-year serial error correlation.
4.1 Effect of AcquirerAffiliation on upgradesand downgrades
As discussed in section 2, the effect of acquirer affiliation will depend on whether the
analyst is reporting on the acquirer or on the target. In Section 4.1.1 we examine reports on the
acquirer. In section 4.1.2 we examine reports on the target.
4.1.1 Acquirer affiliation and upgrades of acquirerstock
In this section, we analyze recommendation changes about acquirer stock issued by
acquirer-affiliated and unaffiliated analysts within 90 days of the M&A transaction. Our first
logistic model tests whether acquirer-affiliated analysts are more likely to upgrade acquirers than
are unaffiliated analysts:
P(upgrade) = A(a + PJAffilAcq + fl2 Value + P6 Days + f3 Size + 34 Experience +
5sFollowing + e)
(Model L1)
where upgrade is a variable that takes on a value of 1 if the analyst upgraded the stock, zero if he
left his rating unchanged, and -1 if he downgraded 1. AffilAcq is a dummy indicating that the
analyst is affiliated with the acquirer. Both the conflict of interest and selection bias hypotheses
predict that AffilAcq should increase the probability of an analyst upgrading acquirer stock. The
variable Value is total consideration the acquirer paid for the transaction, including all cash,
securities, and assumed debt, in billions of dollars.
Our analysis also controls for the following variables (see table 5 for detailed variable
definitions): number of days between the forecast date and the M&A transaction date (Days), the
analyst's experience (Experience), defined as the number of years between the current
recommendation and the analyst's first in I/B/E/S, the market capitalization of the firm that is the
subject of the analyst report (Size), and the total number of analysts following the company
(Following). We henceforth refer to this group of variables as the vector controls. We control
for the number of days between the M&A transaction and analyst report because affiliated
analyst recommendations, if they are optimistically or pessimistically biased, are more likely to
be biased closer to the transaction date. We include the length of the analyst's career because
recent research shows that more experienced analysts are more likely to deviate from consensus
(Hong, Kubik, and Solomon, 2000). Market capitalization (Size) is included to proxy for the
uncertainty about future earnings, which is likely lower for larger firms. Finally, we control for
the total number of analysts following the company because a greater number of analysts
increases the likelihood of herding.
[Table 5]
Results for Model L1 are presented in Table 6, Panel A. We report parameter estimates
and their standard errors as odds ratios. The odds ratio associated with AffilAcq is greater than 1
and statistically significant at the 1% level, constituting strong evidence that acquirer-affiliated
analysts are more optimistic about acquirers.
The magnitude of the odds ratio indicates that
acquirer affiliation increases the odds of an upgrade of the acquirer by a factor of 1.5, which
seems economically large.
[Table 6]
The dummy AffilAcq treats all deals equivalently. However, affiliated analysts' incentive
to issue biased reports may be greater in larger deals, since deal size is correlated with M&A
fees. We thus run Model L2, which is identical to Model 1, except that it includes the interaction
between Value and AffilAcq:
P(upgrade) = A(a + fIAffilAcq + f 2Value + /3Value*AffilAcq + controls)+ E
(Model L2)
We do not use M&A fees directly in the specification because the number of deals for which fee
information is available is very small. Consistent with both the selection bias and conflict of
interest hypotheses, the odds ratio associated with the interaction term is greater than 1 and
statistically significant. Thus, deal size magnifies the effects of acquirer affiliation.
To identify the effects of conflict of interest, we estimates Models L-3 and L-4, which are
identical to models L1 and L-2, except we only include cash deals in the sample. Recall from
section 2 that optimism toward the acquirer on the part of acquirer-affiliated analysts in cash
deals would constitute unambiguous evidence in favor the conflict of interest hypothesis.
Affirming conflict of interest, we find that acquirer affiliation increases the odds that an analyst
will upgrade the acquirer in cash deals, and the effect is both statistically and economically large.
The odds ratio estimate associated with AffilAcq in Model L3 implies that acquirer-affiliation
increases the odds of an upgrade by a factor of more than 2, and this estimate is statistically
significant at the 1% level. Curiously, the effect is stronger for smaller cash deals, as the odds
ratio for the interaction term in model L4 is less than 1. The parameter on this term, however, is
not statistically significant.
As a robustness check, we attempt to control for deal quality by including the marketadjusted return for the 3-day window around the announcement date, and the results, not
100
tabulated, are unchanged. In addition, we partition the sample according to whether the deals
had positive or negative announcement returns and re-run our regressions, without tabulating the
results in the paper. Acquirer affiliation has a qualitatively similar effect on the likelihood of an
upgrade in both samples. Quantitatively, the effect is stronger by an economically but not
statistically significant amount in the negative return sample, which provides some support to the
conflict of interest hypothesis, as it indicates that acquirer affiliated analysts tend to give a "shot
in the arm" to acquirer stock when the deal is not well received.
4.1.2 AcquirerAffiliation and Upgrades of Target Stock
Recall from section 2 that both the selection bias and conflict of interest hypotheses
predict that acquirer-affiliated analysts might produce bearish reports on the target. However,
such reports might also be bullish because deals about targets with good prospects are more
likely to be acceptable to acquirer shareholders. In order to resolve this question, we consider
the sample of recommendation of target stock issued by acquirer-affiliated and unaffiliated
analysts within 90 days of the M&A deal. We estimate logistic models L5 and L6, which are
identical to L2 and L3, except they are estimated using the sample of changes in
recommendations of target stock. The parameter estimates can be found in columns 5 and 6 of
Panel A of Table 6. Neither AffilAcq nor its interaction with Value is statistically significant.
Hence, we fail to find evidence that acquirer affiliation affects upgrades of target stock. Large
standard errors however, indicate low power, most likely due to there being a relatively small
number of only 62 acquirer-affiliated recommendations of target stock in our sample.
4.2 TargetAffiliation and Upgrades
101
Next, we test how target affiliation influences recommendation changes. Section 4.2.1
examines the effect changes in of target recommendations, while section 4.2.1 examines the
effect on changes in acquirer recommendations.
4.2.1 Target Affiliation and upgrades of TargetStock
Recall from section 2 that target-affiliated analysts have an incentive to be optimistic
about the target according to both the conflict of interest and selection hypotheses. To assess the
effect of target affiliation on the odds that an analyst will upgrade a target stock, we estimate
logistic Models L7 and L8. These models are identical to Models L1 and L2, except they use the
target affiliation dummy, AffilTar, instead of AffilAcq, and we estimate them for the sample of
recommendations of target stock issued by target-affiliated and unaffiliated analysts:
P(upgrade) = A (a + PltAffilTar + J 2Value +controls) + e (Model L7)
P(upgrade) = A (a +flPAffilrar +fl2 Value+fl Value*Affilrar +controls) + E (Model L8)
We present the results in Panel B of Table 6, columns 1 and 2. In Model L7, the
coefficient on the target-affiliation dummy is not statistically significant, implying that we
cannot reject the null hypothesis that, on average, target-affiliated analysts are no more likely to
upgrade the target than are unaffiliated analysts. However, in Model L8, which includes the
interaction with Value, the coefficient on Affilrar becomes positive and highly significant. Thus,
target-affiliated analysts tend to upgrade target stocks when the interaction effect is controlled
for. The parameter estimates imply that target affiliation in a $1 billion deal increases the odds
that an analyst upgrades the target by a factor of approximately 1.84, as indicated by the product
of the odds ratios corresponding to Affilrar and ValuexAffilTar.
4.2.2 Target Affiliation and Upgrades ofAcquirer Stock after Stock Deals
102
Recall from Section 2 that under the conflict of interest hypothesis, in stock deals, targetaffiliated analysts should become more optimistic about the acquirer after the exchange ratio is
set. This is because target shareholders and rmnagers benefit from an increase in the acquirer's
price after this date. This kind of behavior is not consistent with selection bias, since analyst
behavior after the exchange ratio has been set is unlikely to affect deal execution ability.
We test this hypothesis by examining whether target-affiliated analysts are more likely to
upgrade acquirer stock after all-stock transactions have been announced. As discussed in section
3, we use the transaction announcement date to proxy for the exchange-rate fixing date because
previous research has found the two dates to correspond 78% of the time (Mitchell, Pulvino and
Stafford, 2002). We estimate the following logistic models on the sample changes in targetaffiliated and unaffiliated recommendations of the acquirer:
P(upgrade)= A(a +PI;Affilrar
+
J 2Value +f3After + controls)+ E (Model L9)
P(upgrade)= A(al +PIAffilTar +- 2Value +f 3After + J 4After*Affilrar + controls)+ E (Model L10)
P(upgrade)=A(cz +PlAffilrar+ fl2 Value +f 3After +P4*After*Affilrar + f 4 *Value*After*Affilrar
+ controls) + E
(Model L 11)
Odds ratios and standard errors corresponding to the parameters in this model can be found in
Panel B of Table 6, columns 4 and 5.
The odds ratio corresponding to AfterxAffilrar
in Model L10 is statistically
indistinguishable from one. This indicates that on average, target-affiliated analysts do not
become more optimistic about acquirers after the approximate date upon which the exchange
ratio has been set. Notice, however, that when we interact AfterxAffilrar with Value in Model
L1 l, we obtain an odds ratio that is significantly greater than 1. This result implies that when
stock deals are large, target-affiliated analysts do in fact become more optimistic about the
103
acquirer after the setting of the exchange ratio, which is consistent only with the conflict of
interest hypothesis. For large deals, the economic magnitude of this effect is large. To quantify
it, we run Model L12, identical to Model L10 except we run it on a sample of recommendations
published around deals valued at more than $2 billion. For such large deals, target affiliation
increases the odds of an upgrade of the acquirer after the setting of the exchange ratio by a factor
of more than 2, as indicated by the product of the odds ratios corresponding to AffilTar and
AfterxAffilrar.
As a robustness check, we attempt to control for deal quality by including the marketadjusted return for the 3-day window around the announcement date. The results, not tabulated,
are unchanged. We also partition the sample according to announcement returns. Estimates of
the coefficients (not tabulated) on After*Affilrar in Model L12 and Value*After*Affilrar are
larger in the positive return sample, but not by statistically or economically amount. We thus
conclude that the affect of target affiliation does not significantly change with deal quality.
5. Ordinary Least Squares Analysis of Long-Term Growth Forecasts
In this section we use regression analysis to determine whether there is an association
between M&A relationships and long-term growth forecasts of analysts affiliated with M&A
advisors. In Section 5.1 we examine the effects of acquirer affiliation on analyst long-term
growth forecasts. In Section 5.2 we examine the effects of target affiliation. For all regressions,
we compute standard errors by clustering by calendar year, ensuring robustness to
heteroskedasticity and arbitrary within-year cross-sectional and serial error correlation. In each
case, the dependent variable, optimism, is defined as the analyst's forecast less the unaffiliated
consensus in the same calendar month. All of the OLS specifications in this section have the
same independent variables as the logistic specifications in Section 4.
104
5.1 Acquirer affiliation and long term growth forecasts
5.1.1 Acquireraffiliation and Reports on the Acquirer
In this section, we analyze long-term growth forecasts about the acquirer issued within 90
days of the M&A announcement date by acquirer-affiliated or unaffiliated analysts. We use
OLS specifications, Models 1 and 2, that have the same independent variables as logistic Models
L1 and L2 presented in Section 4.1. Results are presented in Panel A of Table 7.
Consistent with both the selection bias and conflict of interest hypotheses, the coefficient
on the interaction between AffilAcq and Value is positive and statistically significant, but
economically small. Thus analysts affiliated with acquirers in large deals tend to issue slightly
more optimistic forecasts of acquirer growth. Note that the coefficients on the interaction term
and on the affiliation dummy sum to a value less than zero. Because Value is measured in
billions of dollars, this implies that for a $1 billion deal, acquirer-affiliated analyst growth
forecasts tend to be lower than consensus. For a $10 billion deal, such forecasts are higher than
consensus by less than 1%, a modest amount compared to the mean forecast of about 20%.
[Table 7]
To isolate the effect of conflict of interest from selection bias, we estimate the same
regression specifications on the sample of acquirer-affilaited and unaffiliated forecasts of
acquirer firms issued around all-cash M&A deals. The results are in Table 7 and labeled as
Models 3 and 4. The coefficient on AffilAcq is positive in the expected direction, but with a t
statistic of 1.54, is significant only at the 10% level using a one-sided test. Its economic
significance is also small. The coefficient Value of 0.832 implies that acquirer-affiliated analyst
forecasts of acquirer growth are 0.832 percentage points higher than the unaffiliated consensus, a
modest amount compared to the mean forecast of about 20%. The interaction coefficient has the
105
wrong sign, but is statistically and economically insignificant. Thus far, therefore, we find
evidence that acquirer-affiliation does increase analyst growth forecasts about the acquirer, but
only by modest amounts.
5.1.2 Acquirer-affiliation and growth forecasts for the target
In this section, we analyze acquirer affiliated and unaffiliated analyst forecasts about
target growth. We estimate OLS Models 5 and 6, which have the same independent variables as
L-Models 5 and 6 discussed in Section 4. Results are presented in the last two columns of Table
7, Panel A.
Our estimate of the coefficient on AffilAcq in Model 5 is negative and economically large:
acquirer-affiliated analysts on average issue growth forecasts about the target that are 5.585
percentage points lower than consensus. The figure, however, is statistically insignificant. The
standard error of 6.771 suggests low power, however, which is likely due to the small number of
observations, 482, in this sample. We thus cannot make any strong inferences about the effect of
acquirer affiliation on forecasts of target growth.
4.2 Analysis of target-affilaitedanalysts
5.2.1 Target-affiliationandforecasts of targets
In this section, we restrict our attention to forecasts of target growth issued by targetaffiliated an unaffiliated analysts. We estimate OLS models 7 and 8, which have the same
independent variables as logit models L7 and L8 discussed in Section 4. Results are presented in
the first two columns of Table 7, Panel B. Contrary to both conflict of interest and selection
bias, the coefficient on Affilrar is negative, but it is not statistically significant. With a value of
only -1.194, it is also economically insignificant.
The coefficient on the interaction term in
model 8 is also statistically and economically insignificant. The standard error on AffilTar is also
106
small, having a value of only 1.4% compared to the mean growth forecast of 20%, so our power
to detect bias is high.
We conclude, therefore, that target affiliation does not significantly
influence analyst forecasts about target stock growth.
5.2.2 Target-affiliationandforecasts of acquirersin stock deals
In this section, we study the effect of target affiliation on forecasts of acquirer growth.
We estimate OLS models 9-11, which have the same independent variables as logit Models L9L 11 presented in Section 4. We restrict our attention to target-affiliated and unaffiliated analysts.
Results are presented in the last three columns of Table 7, Panel B. Recall from section 2 that
the conflict of interest hypothesis predicts that target-affiliated analysts should become optimistic
about acquirers after, but not before, the exchange ratio is set. Recall also that selection bias
would not explain such a result. As in Section 4, we use the announcement date as our proxy for
the setting of the exchange ratio.
Consistent with the conflict of interest hypothesis, the coefficient on AfterxAffilr,,ar term
in Model 10 is positive. However, it is not statistically significant. To see whether the effect is
larger for larger transactions, we interact AfterxAffilrar with Value. The point estimate for the
coefficient on the triple interaction is also positive, and thereby consistent with the conflict of
interest hypothesis, but it is not statistically significant either. In both cases, the magnitude of
our point estimates also indicates that the economic significance is small. We thus conclude that
target affiliation has little to no significant effect on forecasts of acquirer growth.
6. Summary and Conclusions
We find that M&A relations significantly impact analyst objectivity. In particular, we
find that acquirer affiliation increases the odds, by a factor of 1.5, that an analyst will upgrade the
acquirer within 90 days of the M&A transaction. Furthermore, we find this effect to be even
107
stronger for all-cash deals, wherein the analyst's opinion of acquirer stock is irrelevant to the
advisor's deal execution ability. Hence selection bias is unlikely to explain this last result,
leaving us to conclude that it is most likely the result of conflict of interest. We believe this
result is important because M&A is the largest source of investment banking fees. Hence it is a
large potential source of conflict of interest, until now largely ignored by academics. In addition,
the M&A context allows us to rule out selection bias as an alternative hypothesis, a feat that has
been somewhat elusive in the equity underwriting context.
We also find that target affiliation significantly increases the odds, by more than a factor
of two, that an analyst will upgrade an acquirer soon after the exchange ratio of a large stock deal
is fixed. This finding also constitutes unambiguous evidence in favor of the conflict of interest
hypothesis.
Optimistic reports about the acquirer after a stock deal can benefit target
shareholders and managers since they now hold acquirer stock. There is no obvious legitimate
reason for a target to select an advisor based on a prioriknowledge that the latter's analyst will
behave in this matter, so this result is unlikely due to selection bias. Our finding that targetaffiliated analysts are more likely to upgrade target stock around M&A transactions also supports
the conflict of interest hypothesis, but we cannot rule out selection bias in this case.
While our results indicate that M&A affiliated analysts do indeed bias their reports in
ways that benefit M&A clients and bankers, they do not necessarily indicate that M&A clients
reward or actively solicit such behavior. It is possible that analysts behave this way because they
mistakenly believe M&A clients will reward them. It is also possible that bankers pressure
analysts to shade their reports to benefit clients without the latter exerting any direct influence.
We fail to find that M&A relations significantly affect analyst long-term growth or nearterm EPS forecasts. In the case of growth forecasts, it is possible that few investors pay much
108
attention to them, and hence analysts achieve little by tainting them.
With regard to EPS
forecasts, analyst incentives are unclear. For example, acquirers may not want EPS forecasts to
be excessively high because they might run the risk of failing to meet earnings expectations. In
addition, biasing EPS forecasts is likely more costly to analysts than biasing other parts of their
reports because EPS forecast accuracy can be evaluated more objectively.
109
References
Agrawal, Anup and Mark Chen, 2004. "Analyst Conflicts and Research Quality."
Unpublished working paper. Culverhouse College of Business, University of
Alabama.
Baker, M. and J. Wurgler, 2002. "Market Timing and Capital Structure," Journal of
Finance57, 1-32.
Barber, Brad Rueven Lehavy and Brett Trueman, 2004. "Comparing the Stock
Recommendation Performance of Investment Banks and Independent Research
Firms." Unpublished working paper. University of California, Los Angeles.
Bartov, Eli Dan Givoly and Carla Hayn, 2002. 'The Rewards to Meeting or Beating
Earnings Expectations." JournalofAccounting and Economics 33, 173-204.
Bradshaw, Mark T., Scott A. Richardson and Richard G. Sloan, 2003. "Pump and Dump:
An Empirical Analysis of the Relation Between Corporate Financing Activities
and Sell-Side Analyst Research." Unpublished working paper. University of
Michigan Business School.
"Chronology of the Merrill Lynch Probe." The Associated Press. May 21, 2002.
Cowen, Amanda, Boris Groysberg and Paul M. Healy, 2003, "What types of Analyst
Firms Make More Optimistic Forecasts?" working paper, Harvard Business
School.
Dechow, Patricia M., Amy P. Hutton and Richard G. Sloan, 2000. "The Relation
Between Analysts' Forecasts of Long-Term Earnings Growth and Stock
Performance Following Equity Offerings," Contemporary Accounting Research
17, pp. 1-32.
Dugar, Amitabh and Siva Nathan, 1995, "The Effect of Investment Banking
Relationships on Financial Analysts' Earnings Forecasts and Investment
Recommendations," ContemporaryAccounting Research 12, 131-160.
Krigman, Laurie, Wayne H. Shaw, and Kent Womack, 2001, "Why Do Firms Switch
Underwriters?" Journalof FinancialEconomics 60, 245-84
Goff, John, 2002. "Wall? What Chinese Wall?" CFO.com, April 22, 2002.
Hong, Harrison, Jeffery Kubik and Amit Solomon, 2000, "Security analysts' career
concerns and herding of earnings forecasts," Rand Journalof Economics 31, 121144.
Iskoz, Sergey, 2002. "Relative Performance and Institutional Reaction of Underwriter
Analyst Recommendations." Unpublished Working Paper. MIT Sloan School of
Management.
110
Kasznik, Ron; and Maureen McNichols, 2002. 'Does Meeting Earnings Expectations
Matter? Evidence from Analyst Forecast Revisions and Share Prices." Journalof
Accounting Research 40, 727-59.
Lin, Hsiou-wie and Maureen F. McNichols, 1998, "Underwriting Relationships,
Analysts' Earnings Forecasts and Investment Recommendations," Journal of
Accounting andEconomics 25, 101-127.
O'Brien, Patricia, Maureen F. McNichols and Hsiou-wie Lin, 2005, "Analyst Impartiality
and Investment Banking Relationships," working Paper, Stanford University.
Ljungqvist, Alexander, Felicia Marston and William J. Wilhelm, 2005, "Competing for
Securities Underwriting Mandates: Banking Relationships and Analyst
Recommendations," JournalofFinance, forthcoming.
Malmendier, Ulrike and Devin Shanthikumar, 2004. "Do Securities Analysts Speak in
Two Tounges?" Unpublished working paper, Stanford University.
Malmendier, Ulrike and Devin Shanthikumar, 2005. "Are Small Investors Naive about
Incentives?" Unpublished working paper, Stanford University.
Michaely, Roni and Kent Womach, 1999, "Conflict of Interest and the Credibility of
Underwriter Analyst Recommendations," Review of Financial Studies 12, 653686.
Mikhail, Michael B., Beverly R. Walther and Richard H. Willis, 2005, "When Security
Analysts Talk Who Listens?" Working Paper, Duke University.
NASD and NYSE, 2005. "Joint Report by NASD and the NYSE on the Operation and
Effectiveness
of Research Analyst Conflict of Interest Rules."
http://www.nasd.com/web/groups/rulesregs/documents/rulesregs/nasdw_01580
3.pdf
Mitchell, Mark, Todd Pulvino and Erik Stafford. "Price Pressure Around Mergers."
Unpublished Working Paper, Northwestern University.
Reingold, Dan and Jennifer Reingold, 2006. Confessions of a Wall Street Analyst. New
York: Harper-Collins.
Scharfstein, David and Jeremy Stein, 1990, "Herd Behavior and Investment," American
Economic Review 80, 465-479.
Trueman, 1994, "Analyst forecasts and herding behavior," Review of FinancialStudies 7,
97-124.
Welch, Iwo, 1996, "Herding Among Securities Analysts." Journal of Financial
Economics 58, 369-396.
111
Womack, Kent, 1996, "Do Brokerage Analysts' Recommendations Have Investment
Value?" Journalof Finance 51, 137-167.
112
I
·
ood
o0.
iF
U
,0)
-a,CU
E
Su)
a)
0CD)
Q.
E
o
t
o
d
co
a)
L•
LL
co
0)4
'CD
E
0)
co
I
aD
ui
Ea
0
a)
o
O
Ea
c
'-C
cr
0
F)
ca
E)
LL
U--
,C
m0
C,)
I
I
Table 1
Deviation from Consensus
Analyst
Affiliated
of
Predictions
Panel A: Predictions Assuming Only the Conflict of Interest Hypothesis
Cash and Stock Deals Pooled
Analyst Report on
Analyst
Affiliation
Acquirer
Target
Acquirer
Optimism
Indeterminate
Target
Indeterminate
Optimism
Panel B: Cases that Distinguish Conflict of Interest from
Selection Bias
Analyst
Affiliation
Analyst
Report on Deal Type
Timing of
Report
Prediction
Conflict of
Interest
Prediction under
Selection Bias
Acquirer
Acquirer
Cash
Anytime
Optimism
No Bias
Target
Acquirer
Stock
Before
Exchange
Ratio Set
Pessimism
Pessimism
Target
Acquirer
Stock
After
Exchange.
Ratio Set
Optimism
Pessimism
114
~cdoCZ
•
n
;- 4C Z
;
X Clt-co6r
I0
r41\
C
Cl
cn
IV \
CO
-
C)
C
C)
(fl 66
-aod)
o* 7 :
cc
c
CO
CO~
\oN~
0)-C
- n-0 I: o
200
C)N C> No\ r-o-
abco
CO00-~
tr 66
c
Cl
mCQ
Cl00
0
G6
00
6~~ccn
03 6
0 cE
0
sE.
z-o
12d
cr
112o
VmA(11
C)
N
C)
CN
Sf)
0
0)
00
Cn
Cl
-
1(2
11 :
C~
-"-)
OCO
In
Sf
ON
44
1
C/T
53
(~5
uCO C)
X2-
C)
•
-oo
Hai
~J~
nedCIO a oo
r
04
a
r
~4-j
0
c UO
a,
CU
C-)
-
m
0
rE- O
C
0
o
S0
0
-4
c
--
a,
r
ac
t
~ON c X~4-4
C~
_f
-. '
csi ac
c-
C)~c
c0cr
5•
E!
_0
- (N
V
CU toO
QI)
_0
or
-0 ccr
C
0
ocn
(Cf
<c
.-C,
acc
r
a)I Lr,
C
¢
-0
"1
(c
Zc
0cd
a:
(.2
rIL
C.
0a
c~r
k-
C
V,
oo
3
3
a
C34
CA :ý I:
U(N
C)C)
c3
C
c,
h
6
o
)
0
0
3
0
0
0
0
o
0
0
r0
o
F--0
C
2ý
00
00
0
0
0
6
6
0
.
o
0
0
0
0
0
0
d
6
S
0
0
d0
0
0
o
C)
C)
Cl)
©
1=4
0
0
o
0
0
03
C)
0
0
6
C)
CTS
C'
0\
6
A
0
6
to
00
00
6
N
00
v
o
S
2;
S
.-
;z
Cl)
cl)
©
0
1-4
V-1
C)
00
d
o'Cl
N
6
6
..
00
6c
to
'-)
C)
Cu
<
6
©
---Cl)
c,
-4 *
r-4
S
en
cuu
o1]
0o
Cl: C)
4c4
C,3
UC)
0
o
m
Cý
m
a'
•.•
C)
en
4-
0
4
0
0
0
A
V
c·.
4-
0
-4
a
r
0
0\to
A
1-
C)
0
e0
Co
Cl1
06
Cl1
0
0
I~
.~
C)t
~0'
~~
C)
-
C)3
0
C. )
1-4
C~) m C)
C)
,*~
*O
C)U
C)o =
c~-d
1-4
CI)
C
-
C)
a
C)
CI
)
-
C)
-
Cd
CU
H
,C~a
C)
CU
H~a
a
C)
1-4
0
~
0 1-4
C.JEa
0)
a)
0)
n
C)
i)
~
0o
Y
OI
r.•
C) o
-,4
C)
d)
-4ý
---
·
C
o6
03
o
o
0C
oC
o
C)
00
6
o
o
o
oo
C)
4o
03
-e
a03
0
1r;
N
0C
N
6
bo
0
C)
C)
o
o
C
d
o
O
4ý
6
a
C)
o0
C)
r-L
4-4
C)
e
o
o
o
C;
00
o
C)J
0i
o
0
c
a
Cl
a
'-)
C)
C0
IN
-in
4-4 OC
I
cn
0
o
C)
C)
©
C)
A
0Cl
0
o
e-
oo
0
T:$
C)IN
0o
H--4
!0
Cc)
e3
00
00
bC
O
o
C)
CIO
c
o
C
©
C)
o0
2
C)
0In
0
OO
t--
o
o
c 3C)3
V
o
03
v-,
P~U
·
• ••
z
©
O°
44ý
Jý
0
-
7 <
C4
C0
3
rj•
~4
C
0-4
C))
-
0(o
di
<n
cC
A03
M3.
0
C)
0
En
ý=
F0
C)
42
ct+-
C
0
.3
FCA~ ct
-
F- O
cr~3
.
4-11-t~
-
C3
r
i
00
o
C)
C)
0c)
-e
ci
·
·
Table 5
Variable Definitions
Variables
Definition
A categorical variable that can take on three values: 1 if the analyst's
Upgrade improved within 90 days of the M&A announcement, 0 if it did not change,
or -1 if it was worsened. Used as dependent variable in ordered logit
analysis.
Individual analyst's long-term growth forecast less unaffiliated consensus.
Optimism Used as the dependent variable in the OLS regressions.
AffilAcq
Dummy indicating the analyst is affiliated with the acquirer's advisor
Affilrar
Dummy indicating the analyst is affiliated with the target's advisor
Value
Total value of consideration paid by the acquirer in the M&A deal
Number of analysts either recommending or issuing a forecast about the
Following firm in the same calendar month
Experience Number of years the analyst's forecasts have been appearing in I/B/E/S
Absolute number of days between an M&A deal's announcement date and
Days
the date of the analyst forecast or recommendation
The market capitalization of the firm that is the subject of the analyst report
Size
as of the last trading day of the calendar month preceding the analyst report
Dummy indicating that the forecast or recommendation is issued after the
After
M&A transaction announcement date.
119
Table 6
Effect of M&A Affiliation on Analyst Upgrades and Downgrades
Ordered Logistic Regression Analysis
We model the probability that an analyst will improve a stock recommendation around the
time of an M&A deal as a function of the analyst's affiliation, the value of the deal, various
interaction terms, and various control variables given in Table 5. Panel A presents odds ratio
estimates and standard errors for various subsamples of acquirer-affiliated and unaffiliated
analyst upgrades and downgrades. Panel B presents the odds ratios for various subsamples
of target-affilaited and unaffiliated analyst upgrades and downgrades. Subsamples are
partitioned according to whether the acquirer or target is the subject of the report, and
whether the deal is cash or stock. Standard errors are computed by clustering on calendar
year, making them robust to heteroskedasticity and arbitrary cross-sectional and intra-year
serial correlation. The indicated prediction for the odds ratio estimated is under the conflict
of interest hypothesis.
I
Panel A: Measuring the Effect of Acquirer Affiliation
Reporting on Acquirer
All Deals
Cash Deals
Prediction Model L1
Test Variables
AffilA r
>1
Value*AffilA,,
>1
Value
Controls
Size
Expr
Following
Days
Model L2
Model L3
Model L4
2.122***
2.143***
(0.252)
1.005*
(0.003)
1.288*
(0.184)
1.141**
(0.074)
1.004
(0.003)
(0.343)
0.988
(0.078)
1.097**
(0.049)
1.084***
1.083***
(0.012)
(0.012)
0.956
(0.103)
0.950***
0.956
(0.019)
1.017
(0.102)
(0.145).
0.950***
(0.013)
1.528***
(0.165)
(0.013)
1.001**
(0.001)
Observations
13007
Pseudo R2
0.005
Robust standard errors in parentheses
1.096**
(0.048)
Re:porting on Target
All Deals
Prediction Model L5 Model L6
<1
1.096
(0.507)
<1
1.372
(0.688)
0.989
(0.008)
1.011**
1.012"*
(0.005)
(0.005)
1.035
(0.040)
1.069
(0.081)
(0.040)
0.966*
1.045**
(0.019)
1.017
(0.145)
0.966*
0.972
1.071
(0.082)
0.972
(0.017)
(0.01.7)
1.001**
'1.001.
1.001
(0.018)
0.992***
(0.018)
0.992***
(0.001)
(0.001)
(0.001)
13007
0.005
(9.001)
5814
2278
0.014
2278
0.014
1.045**
0.005
* significant at 10%; ** significant at 5%; *** significant at 1%
120
5814
0.005
1.030
cl
A
(d
0C
u
co
ci)
0,
-
' ci
0fl
Oc CD fn
C
6
ciC)
N "-"
01
O\ It
"•
66
al
E]
ci
..
0
ci)
00
*00 ril)
6
C14
C) 00
6
o
-
'ia
2
o,-,
u
"
C
*
Aa00
oc- ri
agg-
"
c:
m
40
CI
~ OCOm
co
(
r
a
oo\o-Z
-
•" ,
C
,C
=
o
ao
,c
0
c,Z )o
0
" o
Coz C-A
cc)
0O
I)
c.<)
0l 0
0O
ci
2
00
A
ci)~
- -- 0 ..
A
:i
*
o -_
Cozoo
-.-
on
O--o
0
,c
--:'1i .- ON
CcO ,-N
* o
--
c
,c
o
ci
oI,
ci)
ci)
0~
c/
F-
0
on
0
0
0
*A
A
C)
6
6c
6-
C)h
ci)
C
y
(`\
\
~N
*d
4m
\
30\HC
0
cD
00
C
ci)
00o
00
cC
OIC
a, >-, <
0e
*0
<:
>
cci
<::-
<:
1
--.
:>I
__
>-
<:
,e
C
011
0
X "-o
C)
r.. u..,
C-p
o
Table 7
OLS Analysis of the Effect of M&A Affiliation on Long-Term Growth Forecasts
We model analyst deviation from consensus growth forecast as a linear function of the analyst's
affiliation with an M&A advisor, the value of the deal, various interaction terms, and various control
variables defined in Table 5. Panel A presents OLS estimates and standard errors of model
parameters for various subsamples of acquirer-affiliated and unaffiliated analyst forecasts. Panel B
presensts the estimates and standard errors for various subsamples of target-affilaited and
unaffiliated analyst forecasts. Subsamples are partitioned according to whether the acquirer or target
is the subject of the report, and whether the deal is cash or stock. Standard errors are computed by
clustering on calendar year, making them robust to cross-sectional and intra-year serial correlation.
"Prediction" is the expected sign of a coefficient estimate under the conflict of interest hypothesis.
Panel A: Measuring Effect of Acquirer Af filiation
Forecasting Acquirer
All Deals
Forecasting Target
Cash Deals
Prediction
Model 1
Model 2
Model 3
+
0.045
(0.575)
-0.136
(0.674)
0.832
1.069
(0.542)
(0.711)
Model 4
Prediction
All Deals
Model 5
Model 6
'rTest Variables
AffilA,
Value*AffilAca
Value
+
-5.677
(6.622)
-0.276
(0.492)
0.105**
-6.413
(7.476)
0.061
(0.081)
(0.042)
-0.005
(0.011)
0.050
0.056
(0.155)
(0.152)
0.047**
(0.017)
0.043**
(0.017)
-0.082
(0.130)
-0.170
-0.170.
-0.604*
(0. t145)
0.936***
(0.291)
-0.012
::0.936***
0.958**,
(0,.144):
0.960**
(0,311)
1.473**
(0.311)
(0.31 1)
-0.034**
(0.010)
(0.0 10)
-0.001
-0.034**
(0.012)
(0.470)
0.077
(0.057)
0.006
(0.016)
482
0.011
-0.582
(0.325)
1.461"*
(0.470)
0.075
(0.058)
0.005
(0.016)
-0.005
(0.011)
Controls
Size
-0.081
(0.130)
Expr
Following
Days
-0.001
(0.004)
...(0.291)
-0.012
(0.004)
(0.012)
0.001
(0.006)
Observations
6769
3023
6769
R-squared
0.003
0.003
0.007
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
122
0.001
(0.006)
3023
0.007
482
0.011
Panel B: Measuring the Effect of Target Affiliation
Forecasting Target
_
Forecasting Acquirer, All-Stock Deals
_
Prediction
Model 7
Model 8
Prediction
Model 9
Model 10
Model 11
AffilTAR
+
-1.214
(1.296)
?
-0.279
(1.259)
-0.481
(2.180)
-0.481
(2.181)
Value*AffilrAR
+
-1.648
(1.652)
0.034
(0.019)
0.374
(2.960)
0.237
(3.120)
0.031
(0.077)
-0.016
(0.012)
0.461
(0.531)
Test Variables
After*AffilA,
I
Value*After*AffilTAR
Value
0.045**
(0.016)
0.038*
(0.018)
-0.015
(0.013)
(0.378)
1.054.
(0.165)
(0.784)
(0.464)
After
Controis ::
Size
Exper
-0.627:
-0.039
0.111*
0.112*
(0::.056)
0.003
0.003
Days
(9.017)
Observations
487
487
0.006
R-sciuared
0.006
Robust standard errors in parentheses, clustered by calendar year
Following
(0.056).
.(0.017)
* significant at 10%; ** significant at 5%; *** significant at 1%
123
0.881*
-0.006
(0.018)
-0.002
(4.006)
2521
0.002
-0.015
(0.012)
0.461
(0.531)
-0.034
-0.034
(0.165)
(0.165)
0.862"..
0.861"*:
. (0.456):: (0.456)
-0.007
(0.018)
-0.002..
:.:-0.007
(0.018)
-0.002.
(0.0o6)
(9.0906)
2521
0.003
2521
0.003